To begin searching the abstracts, please use the search feature above.

Digital Poster - Thursday
Weekend and Oral

Digital Poster (no CME credit)

Monday Digital Poster Tuesday Digital Poster Wednesday Digital Poster

Thursday Digital Poster (No CME Credit)

Acquisition, Reconstruction & Analysis

08:15
4376 - 4545
09:15
4546 - 4718
13:45
4719 - 4893

Contrast Mechanisms

14:45
4894 - 5068

Using MRI to Measure Numbers Outside of the Brain

Exhibition Hall
Thursday 8:15 - 9:15
 Acquisition, Reconstruction & Analysis

4376
Computer 1
Fat/Water Separation and T1 and T2 Quantification Using MRF with a Rosette Trajectory in the Heart and Liver
Yuchi Liu1, Jesse Hamilton1, Mark Griswold1,2, and Nicole Seiberlich1,2

1Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 2Department of Radiology, University Hospitals Cleveland Medical Center, Cleveland, OH, United States

Cardiac Magnetic Resonance Fingerprinting (cMRF) has recently been introduced for simultaneous T1 and T2 quantification in the myocardium. One important feature of the MRF framework is the potential to measure multiple tissue properties beyond T1 and T2. Here we propose an approach for simultaneous fat imaging and T1 and T2 quantification based on the cMRF framework with a rosette trajectory. The accuracy in T1 and T2 measurements and the efficacy in water-fat separation were demonstrated in the ISMRM/NIST system phantom and a multi-compartment water/oil phantom, respectively. Preliminary results in the heart and liver in healthy subjects are also shown.

4377
Computer 2
Mapping T1, T2, and proton density fat fraction of the liver using MR Fingerprinting with three-point DIXON and 6-peak fat model
Daiki Tamada1, Hiroshi Onishi1, and Utaroh Motosugi1

1Department of Radiology, University of Yamanashi, Chuo, Japan

An MR Fingerprinting (MRF) simultaneously combining the three-point DIXON (3P-DIXON) method for the fatty liver was proposed. The MRF-FISP sequence with multi-TR/TE/flip angle was developed. The six-peak fat model was used to calculate a dictionary for the MRF. Template matching using the acquired signal evolutions and the rough fat fraction map estimated by 3P-DIXON provided quantification of T1, T2, and fat fraction. The phantom and volunteer studies demonstrated the feasibility of our study.

4378
Computer 3
Three-Dimensional, Free-Breathing Magnetic Resonance Fingerprinting for Whole-Liver Coverage
Kathleen M. Ropella-Panagis1, Yong Chen2, Yun Jiang1, Jesse Hamilton3, Wei-ching Lo3, Dan Ma1, Mark A. Griswold1, Nicole Seiberlich3, and Vikas Gulani1

1Department of Radiology, Case Western Reserve University, Cleveland, OH, United States, 2Department of Radiology, University of North Carolina at Chapel Hill, Chapel Hill, NC, United States, 3Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States

In this proof-of-concept study, a 3D free-breathing abdominal MR Fingerprinting sequence is applied to the abdomen. Full-liver coverage with spatial resolution of 1.6×1.6×5 mm3 is attained in 8 minute 40 seconds.

4379
Computer 4
Free-Breathing Liver T1 and Fat Mapping Using a Golden-Angle-Ordered Variable Flip Angle Stack-of-Radial Sequence
Le Zhang1, Tess Armstrong1,2, and Holden H. Wu1,2,3

1Radiological Sciences, University of California, Los Angeles, Los Angeles, CA, United States, 2Physics and Biology in Medicine Interdepartmental Program, University of California, Los Angeles, Los Angeles, CA, United States, 3Department of Bioengineering, University of California, Los Angeles, Los Angeles, CA, United States

Measurement of T1 and proton-density fat fraction (PDFF) in the liver can provide information about fibrosis and steatosis, respectively. Existing Cartesian acquisition schemes generally require breath-holding, which limits spatial coverage and may be difficult for sick, elderly or pediatric patients. In this study, we propose a golden-angle-ordered (GA) 3D stack-of-radial variable-flip-angle (VFA) sequence that can map T1 and PDFF simultaneously with close to full liver coverage under five minutes during free-breathing. Pilot studies in phantom and healthy subjects demonstrate feasibility and show good measurement repeatability.

4380
Computer 5
DeepKidney: Deep segmentation of MR images for automated glomerular function quantification in heterogeneous pediatric patients
Edgar A. Rios Piedra1, Morteza Mardani1, Ukash Nakarmi1, Joseph Y. Cheng1, and Shreyas S. Vasanawala1

1Department of Radiology, Department of Electrical Engineering, Stanford University, Palo Alto, CA, United States

Automated segmentation of kidneys and their sub-components is a challenging problem, particularly in pediatric patients and in the presence of a pathology or some anatomical deformation. We present a segmentation framework using a multimodal U-Net that allows for the automated segmentation of the multiple kidney components as well as a functional evaluation of the glomerular filtration rate. Results achieve an average Dice similarity coefficient of 0.912, 0.853, and 0.917 for kidney cortex, medulla, and collector system, respectively.

4381
Computer 6
Field Map Estimation from Magnitude-Based Water-Fat Separation
Alexandre Triay Bagur1, Chloe Hutton1, Benjamin Irving1, Michael L. Gyngell1, Matthew D. Robson1, and Michael Brady1

1Perspectum Diagnostics Ltd, Oxford, United Kingdom

Complex-based MRI chemical-shift encoded water-fat separation depends on accurate field map convergence, which is often mitigated with spatial regularization. This is prone to error propagation and over-smoothing of fat-fraction maps. Magnitude-based separation circumvents field mapping but is reportedly limited in fat-fraction range (0-50%). We have recently presented MAGO, a magnitude-based method that resolves this water-fat ambiguity. In this study, we compare MAGO to state-of-the-art fat-fraction quantification on N=150 volunteers, and we expand the method for field map calculation using previously estimated water and fat images. MAGO is comparable to regularized hybrid-based decomposition and shows promise in higher field inhomogeneity regimes.

4382
Computer 7
Phase Correction for Abdominal Quantitative Susceptibility Mapping with Bipolar Readout Gradients Sequence
Rui Tong1, Jianqi Li1, Xu Yan2, and Yi Wang3

1Shanghai Key Laboratory of Magnetic Resonance, Shanghai, China, 2MR Collaboration NE Asia, Siemens Healthcare, Shanghai, China, Shanghai, China, 3Department of Radiology, Weill Medical College of Cornell University, New York, New York, USA, New York, NY, United States

Bipolar acquisition in abdominal multi-echo quantitative susceptibility mapping (QSM) could reduce echo-spacing and total scan time. However, the bipolar acquisition introduces phase error between odd and even echoes. A phase correction method in image domain was proposed to address this problem. We demonstrated the feasibility of generating a quantitative susceptibility map in human abdomen using bipolar multi-echo GRE sequence. Quantification analysis showed an excellent agreement between bipolar and unipolar methods.

4383
Computer 8
Quantifying liver function using artificial neural networks to estimate gadoxetic-acid uptake rate in temporally sparse gadoxetic-acid enhanced MRI
Josiah John Simeth1,2 and Yue Cao1,2

1Biomedical Engineering, University of Michigan, Ann Arbor, MI, United States, 2Radiation Oncology, University of Michigan, Ann Arbor, MI, United States

Though methods exist for quantifying regional liver function from dynamic gadoxetic-acid enhanced (DGE) MRI, errors are introduced when using the clinically typical temporally sparse acquisition scheme (6 volumes over 20 minutes) relative to a temporally dense dynamic acquisition (volumes every 5-10 sec over a similar period). This motivates a data driven approach. An artificial neural network (ANN) was trained to reproduce the results of the fully characterized analysis using only the restricted dataset. Across the patients evaluated the ANN solution resulted in lower mean and median WMAPE, as well as a reduction in MSE in most cases.

4384
Computer 9
multiMap: A Gradient Spoiled Sequence for Quantitation of B1, B0, T1, T2, T2*, and Fat Fraction
Nicholas Dwork1, Adam B. Kerr2, Ethan M. I. Johnson3, and John M. Pauly1

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Stanford Center for Cognitive and Neurobiological Imaging, Stanford University, Stanford, CA, United States, 3Biomedical Engineering, Northwestern University, Evanston, IL, United States

multiMap is a single sequence that combines standard techniques for measuring several quantities of interest: B0, B1, T1, T2, T2*, and Fat Fraction.  

4385
Computer 10
Toward 3D Free-breathing Cardiac Magnetic Resonance Fingerprinting
Gastao Cruz1, Olivier Jaubert1, Torben Schneider2, Aurelien Bustin1, René M. Botnar1, and Claudia Prieto1

1School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, 2Philips Healthcare, Guildford, United Kingdom

Magnetic Resonance Fingerprinting (MRF) has been introduced to simultaneously estimate multiple quantitative parameters but mainly applied to static organs. Recently the feasibility of 2D triggered cardiac MRF (cMRF) under breath-hold has been demonstrated and provides single slice simultaneous T1 and T2 maps. However, 2D cMRF provides insufficient coverage of the heart. Here we sought to develop a free-breathing 3D triggered cMRF sequence. Respiratory bellows drive an autofocus algorithm that is used to perform translation correction of respiratory motion followed by a low rank MRF reconstruction. The proposed 3D cMRF approach was evaluated in three healthy subjects, demonstrating considerable improvements in parametric maps when compared to no motion correction.

4386
Computer 11
Confounding Factors in Breast Magnetic Resonance Fingerprinting: B1+, Slice Profile and Diffusion Effects
Teresa Nolte1,2, Mariya Doneva3, Thomas Amthor3, Peter Koken3, Nicolas Gross-Weege2, Tianyu Han2, Hannah Scholten2, and Volkmar Schulz2

1Philips Research Europe, Eindhoven, Netherlands, 2Physics of Molecular Imaging Systems, RWTH Aachen University, Aachen, Germany, 3Philips Research Europe, Hamburg, Germany

In this study, we evaluate the effect of three potentially confounding factors (B1+ inhomogeneity, slice profile, diffusion) on the outcome of 2D Magnetic Resonance Fingerprinting measurements in the female breast for six healthy volunteers. Each of these factors was included into an MRF dictionary and matching results were compared to a reference dictionary that excluded the correction. For the given MRF sequence, both B1+ inhomogeneity and slice profile correction affected the quantitative relaxation times in the female breast, whereas this was not the case for diffusion.

4387
Computer 12
A Protocol for Comprehensive Quantitative 3D Ultrashort Echo Time (UTE) Cones MR Imaging of the Knee Joint with Motion Correction
Mei Wu1,2, Wei Zhao1, Jonathan Lee1, Lidi Wan1, Saeed Jerban1, Eric Y Chang1,3, Jiang Du1, and Yajun Ma1

1Department of Radiology, University of California, San Diego, San Diego, CA, United States, 2Department of Radiology, Guangzhou First People’s Hospital, School of Medicine, South China University of Technology, Guangzhou, China, 3Radiology Service, VA San Diego Healthcare System, San Diego, CA, United States

We propose a protocol for comprehensive quantitative 3D UTE-Cones imaging of the knee joint with motion correction. The protocol includes 3D UTE-Cones actual flip angle imaging (UTE-Cones-AFI) for T1 measurement, UTE-Cones with variable TEs for T2* measurement, UTE-Cones with adiabatic T preparation for AdiabT measurement, and UTE-Cones-MT for measuring MTR and modeling of macromolecular fraction (f) for various knee joint tissues including the cartilage, menisci, ligaments, tendons and muscle. An elastix motion registration method was used for motion correction. In our study, three knee specimens and 15 volunteers were evaluated. Mean and standard deviation of the measurements for various knee joint tissues are reported.  

4388
Computer 13
Free-running 3D Radial Myocardial T1 Mapping using Self-Calibrating GRAPPA Operator Gridding for Accelerated Iterative Reconstruction
Haikun Qi1, Aurelien Bustin1, Olivier Jaubert1, René Botnar1, and Claudia Prieto1

1School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom

Free-running 3D radial (kooshball) sampling is suitable for fast and self-navigated whole-heart cardiovascular imaging. However, iterative undersampled 3D radial reconstruction requires computational demanding gridding/regridding steps in each iteration, which leads to long reconstruction time and may limit the applications of this imaging strategy. In this work, we investigate the feasibility of accelerating iterative reconstruction for a free-running 3D myocardial T1 mapping sequence using GRAPPA Operator Gridding (GROG)-based pre-reconstruction interpolation. Image quality and T1 estimation accuracy of the accelerated GROG-based reconstruction were compared with conventional non-uniform FFT (NUFFT)-based reconstruction in a standardized phantom and five healthy subjects. 

4389
Computer 14
Efficient Quantitative Susceptibility Mapping of Popliteal Artery Wall
Yan Wen1,2, Thanh Nguyen2, Ajay Gupta2, and Yi Wang1,2

1Meinig School of Biomedical Engineering, Cornell University, new york, NY, United States, 2Radiology, Weill Cornell Medicine, New York, NY, United States

The objective of this study was to develop and optimize the pulse sequence and post-processing for an efficient and high quality QSM of the popliteal artery wall. We showed that high quality QSM could be achieved in 4 minutes without the need for cardiac gating.

4390
Computer 15
Development of a 3D UTE MP2RAGE sequence for mouse pulmonary T1 mapping at 7T
Thibaut L Faller1, Colleen R Cardiet1, Wilfried Souleyreau2, Lin Cooley2, Andreas Bikfalvi2, Baudouin Denis de Senneville3, Sylvain Miraux1, and Emeline J Ribot1

1CNRS - Univ. Bordeaux, CRMSB UMR 5536, Bordeaux Cedex, France, 2INSERM, LAMC INSERM U1029, Pessac, France, 3CNRS - Univ. Bordeaux, IMB UMR5251, Talence, France

The 3D Magnetization Prepared 2 Gradient Echo (MP2RAGE) sequence is very useful to obtain high contrasts between brain tissues and between metastases and the surrounding healthy brain at high clinical magnetic fields (≥3T). In order to apply this sequence for the detection and T1 mapping of lung metastases in mice at 7T, major modifications were done. We developed an ultra-short echo time (UTE) MP2RAGE sequence by replacing the Cartesian encoding by a radial one. This encoding enables (i) to shorten echo time to less than 0.1ms and consequently obtain lung T1 maps; and (ii) to track respiration motion through a self-gating strategy to evaluate the displacements of the metastases due to breathing.

4391
Computer 16
Model Based Analysis of Complex Difference Images for Measuring Diameters and Velocities of Penetrating Arteries
Xiaopeng Zong1 and Weili Lin1

1Department of Radiology and BRIC, Univeristy of North Carolina at Chapel Hill, Chapel Hill, NC, United States

Pathological changes of penetrating arteries (PAs) may be an important contributing factor of cerebral small vessel disease (SVD).  Measurement of PA flow velocity and diameter with phase contrast (PC) MRI remains challenging due to the presence of strong partial volume effects.  Here we propose model-based analysis of complex difference (MBAC) images to quantify diameter and velocity of PAs.  We demonstrated the accuracy of the MBAC method with simulation and phantom studies.  In vivo PA diameter and velocity were obtained for the first time.  The MBAC method may serve as a useful tool for understanding the etiopathogenesis of SVD.

4392
Computer 17
MR Relaxivity Mapping using multi-dimensional integrated (MDI) complex signal ratio
Yongquan Ye1 and Jingyuan Lyu1

1UIH America, Inc., Houston, TX, United States

A novel relaxivity mapping method for MR transverse relaxivity mapping (e.g. T2*) is proposed and demonstrated. By extracting an overall complex signal ratio by means of multi-dimensional integration (MDI) , our method offers significantly improved SNR and homogeneous parametric mappings. With MDI, no explicit multi-channel combination operation is required, and calculation efficiency is extremely high for inline calculation.

4393
Computer 18
Spine Quantitative Susceptibility Mapping Using In-Phase Echoes to Initialize the Nonconvex Optimization Problem of Fat-Water Separation (R2*-IDEAL)
Yihao Guo1,2, Zhe Liu2,3, Yan Wen2,3, Pascal Spincemaille2, Honglei Zhang2, Ramin Jafari2, Shun Zhang2, Sarah Eskreis-Winkler2, Kelly M. Gillen2, Yanqiu Feng1, and Yi Wang2,3

1Guangdong Provincial Key Laboratory of Medical Image Processing, School of Biomedical Engineering, Southern Medical University, Guangzhou, China, 2Department of Radiology, Weill Medical College of Cornell University, New York, NY, United States, 3Department of Biomedical Engineering, Cornell University, Ithaca, NY, United States

This work aims to investigate the initializations of R2* and field maps in R2*-IDRAL for developing a robust quantitative susceptibility mapping (QSM) in the spine. A 3D multi-echo GRE sequence was implemented to acquire out-phase and in-phase (IP) echoes in 10 subjects. The R2* and background field maps estimated by fitting the magnitude and phase of IP echoes were used to initialize R2*-IDEAL to obtain final R2*, field, water, and fat maps. The final field map was further processed to generate QSM. The results demonstrated that IP initializations of R2* and field in R2*-IDEAL provide robust QSM of the spine.

4394
Computer 19
Quantitative Transport Mapping (QTM): a new AIF-free perfusion technique to distinguish malignant and benign breast lesions
Sarah Eskreis-Winkler1,2, Natsuko Onishi2, Liangdong Zhou3, Pascal Spincemaille1, Ramin Jafari3, Meredith Sadinski2, Elizabeth J. Sutton2, Elizabeth A. Morris2, and Yi Wang1

1Weill Cornell Medicine, New York, NY, United States, 2Memorial Sloan Kettering Cancer Center, New York, NY, United States, 3Cornell University, Ithaca, NY, United States

Quantitative perfusion imaging is challenging in the breast because the requisite arterial input function (AIF) is difficult to measure given the lack of large-caliber feeding arteries.  To overcome this problem, we show that quantitative transport mapping (QTM), a new AIF-free perfusion model, is not only technically feasible in the breast, but has the potential to better distinguish malignant from benign breast lesions compared to conventional perfusion modeling.

4395
Computer 20
Texture analysis of multi-phase magnetic resonance images to discriminate expression of Ki67 in hepatocellular carcinoma
Yueming Li1 and Chuan Yan1

1Radiology, First Affiliated Hospital of Fujian Medical University, Fuzhou, China

Aims: This study aimed to determine whether texture analysis of preoperative magnetic resonance images could predict expression of Ki67 in hepatocellular carcinoma(HCC).

Methods: 83 patients confirmed HCC were included. Texture analysis on 3.0 Tesla MR Unit included histogram, co-occurrence matrix, run-length matrix, gradient, auto-regressive model, and wavelet transform features as calculated by MaZda software.

Results: HCC with higher Ki67 label index tend to display a lower differentiation pattern. Larger tumors usually had higher Ki67 label index. Texture parameters generated from arterial phase imaging was the most frequently significant correlation.

Conclusions: Texture analysis could be used to discriminate Ki67proliferation status in HCC.


4396
Computer 21
Free-Breathing 3D T1 Mapping of the Whole-Heart Using Low-Rank Tensor Modeling
Paul Kyu Han1, Debra E. Horng1, Yoann Petibon1, Jinsong Ouyang1, Nathaniel Alpert1, Georges El Fakhri1, and Chao Ma1

1Gordon Center for Medical Imaging, Massachusetts General Hospital and Harvard Medical School, Boston, MA, United States

T1 of the myocardium is an emerging quantitative biomarker for a variety of heart diseases. However, T1 mapping is challenging in the heart due to the cardiac and respiratory motion. To overcome this issue, T1 mapping methods utilizing ECG triggering with breath-hold or respiratory control using bellow or navigators have been developed in the past. Recently, low-rank tensor-based approach has been proposed for MR to enable image reconstruction at extremely high acceleration factors, and has demonstrated promising results. This work presents a low-rank tensor-based approach for free-breathing 3D T1 mapping of the whole-heart.

4397
Computer 22
Diagnostic performance of chemical shift in/opposed phase (IOP) and fat-fraction to evaluate the presence of intra-tumoral fat in HCC
Kritisha Rajlawot1, Jing Zhou2, Churong Lin1, Sichi Kuang1, Jingbiao Chen1, Yao Zhang1, Hao Yang1, Ying Deng1, Bingjun He1, Diego Hernando3, Jin Wang1, and Scott B Reeder3

1Department of Radiology, The third affiliated hospital of Sun Yat-sen University, Guangzhou, China, 2The third Affiliated Hospital, Sun Yat-Sen University, Guangzhou, China, 3Departments of Radiology and Medical Physics, University of Wisconsin, Madison, W, University of Wisconsin, Madison, WI, Madison, WI, United States

Hepatocellular carcinoma (HCC) is the most common primary malignancy of the liver. Previous studies have showed that the presence of intra-tumoral fat has more favorable prognosis of HCC. We aim to compare the diagnostic accuracy of chemical shift encoded fat-fraction at three different flip angles (FA) 3°, 8° and 9°using IDEAL-IQ, in comparison to chemical shift imaging (IOP) to evaluate intra-tumoral fat in HCC.Our results showed higher positive rates of detection of intra-tumoral fat in HCC with IOP (100%) compared to IDEAL IQ at FA 3° (44.4%), FA 8° (55.6%) and FA 9° (88.9%).

4398
Computer 23
DeepBLESS: learning inverse Bloch equations for rapid prediction of myocardial relaxation parameters
Jiaxin Shao1, Vahid Ghodrati 1, Kim-Lien Nguyen2,3, and Peng Hu1,4

1Department of Radiological Sciences, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, CA, United States, 2Department of Medicine, Division of Cardiology, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, CA, United States, 3Division of Cardiology, Veterans Affairs Greater Los Angeles Healthcare System, Los Angeles, CA, United States, 4Biomedical Physics Inter-Departmental Graduate Program, University of California, Los Angeles, Los Angeles, CA, United States

Bloch equation simulation provides accurate estimation of soft tissue relaxation parameters for many applications. To speed up using Bloch equation for relaxation parameter estimation, we propose a general approach - deep learning with Bloch equation simulations (DeepBLESS) - to learn inverse Bloch equation for rapid myocardial relaxation parameter prediction. Using the Modified Look-Locker inversion recovery (MOLLI) sequence and a self-designed simultaneous radial T1 and T2 mapping sequence as examples, we demonstrated that DeepBLESS was adaptive to heart rate variation with good estimation accuracy and precision while reducing the inline computation time compared to the conventional Bloch-equation-based approaches.


Quantitative Mapping of the Brain

Exhibition Hall
Thursday 8:15 - 9:15
 Acquisition, Reconstruction & Analysis

4399
Computer 26
Direct Myelin Volume Fraction Mapping with Correction for Magnetization Transfer and Diffusion Effects Using a Four-pool White Matter Model
Zhe Wu1, Ilana R. Leppert1, and David A. Rudko1,2

1Department of Neurology and Neurosurgery, Montreal Neurological Institute and Hospital, McGill University, Montreal, QC, Canada, 2Department of Biomedical Engineering, McGill University, Montreal, QC, Canada

We propose an accelerated myelin water fraction (MWF) imaging technique that employs wave encoding combined with double inversion-recovery weighting (wave-CAIPI ViSTa) for rapid human brain myelin volume fraction (MVF) measurement. In addition, we investigate the contributions of both magnetization transfer and diffusion to the wave-CAIPI ViSTa MWF signal. A four-pool white matter model and extended phase graph framework are applied for magnetization transfer and diffusion modeling. The proposed MVF mapping method is a promising candidate for measuring white matter properties in human demyelinating conditions with limited dependence on model fitting parameters.

4400
Computer 27
Accelerating Multi-Echo GRASE with CAIPIRINHA for Fast and High-Resolution Myelin Water Imaging
Gian Franco Piredda1,2,3, Tom Hilbert1,2,3, Erick J. Canales-Rodríguez2, Marco Pizzolato3, Reto Meuli2, Josef Pfeuffer4, Jean-Philippe Thiran2,3, and Tobias Kober1,2,3

1Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, 2Department of Radiology, Centre Hospitalier Universitaire Vaudois (CHUV), Lausanne, Switzerland, 3LTS5, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, 4Application Development, Siemens Healthcare GmbH, Erlangen, Germany

Impaired myelin plays a central role in a wide range of degenerative brain diseases. A method for non-invasive and in vivo assessment of myelin content within clinically acceptable acquisition times is thus desirable. In this work, a 3D multi-echo gradient and spin-echo (GRASE) sequence was accelerated with CAIPIRINHA to achieve high-resolution and whole-brain myelin imaging in less than ten minutes. Myelin water fraction (MWF) maps were derived from multi-echo GRASE data in a cohort of healthy subjects and values proved to be consistent with MWF maps computed from a conventional multi-echo spin-echo acquisition.

4401
Computer 28
A fast, joint sparsity constraint algorithm for improved myelin water fraction mapping
Martijn Nagtegaal1, Burkhard Mädler2, Thomas Amthor3, Peter Koken3, and Mariya Doneva3

1TU Berlin, Berlin, Germany, 2Philips, Hamburg, Germany, 3Philips Research Europe, Hamburg, Germany

A new method for myelin water fraction mapping from multi-echo spin echo data using a joint sparsity constraint is proposed, which is faster than previously proposed methods. This method is based on the assumption that the T2 spectrum is sparse and consists of a common small set of discrete relaxation times for all voxels. The method finds an estimation of the flip angle inhomogeneity map from the data itself, to remove the bias caused by B1 inhomogeneities. The proposed method is compared to state of the art MWF approaches in 3T brain measurements.

4402
Computer 29
Non-negative least squares fitting of multi-exponential T2 decay data: Are we able to accurately measure the fraction of myelin water?
Vanessa Wiggermann1,2,3, Irene M Vavasour3,4, Enedino Hernandez-Torres2,3, Gunther Helms5, Alexander L MacKay1,3,4, and Alexander Rauscher1,2,3,4

1Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada, 2Pediatrics, University of British Columbia, Vancouver, BC, Canada, 3UBC MRI Research Centre, University of British Columbia, Vancouver, BC, Canada, 4Radiology, University of British Columbia, Vancouver, BC, Canada, 5Lund University, Lund, Sweden

The ability to determine the myelin water fraction (MWF) in vivo is essential to assessments of neurodevelopmental myelination and myelin damage in neurodegenerative diseases. The analysis of multi-exponential T2 decay data relies on the non-negative-least-squares (NNLS) fitting, which may be sensitive to the chosen fitting parameters. We performed simulations to explore the outcomes of NNLS under different parameter selection. The lowest allowed T2 was found to have the largest effect on correctly estimating the T2 of different water pools as well as the MWF. Lower refocusing FAs led to further underestimation of the MWF.

4403
Computer 30
Myelin Water Fraction Estimation using Small-Tip Fast Recovery MRI
Steven T. Whitaker1, Gopal Nataraj2, Mingjie Gao1, Jon-Fredrik Nielsen3, and Jeffrey A. Fessler1

1Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI, United States, 2Radiology, Memorial Sloan Kettering Cancer Center, New York, NY, United States, 3Biomedical Engineering, University of Michigan, Ann Arbor, MI, United States

Myelin water fraction (MWF) is a good biomarker for myelin content. Traditional methods for acquiring MWF maps require long scan times. Recent work has estimated MWF from faster steady-state scans. In this work, we propose to acquire MWF maps from an optimized set of small-tip fast recovery (STFR) scans that can exploit resonance frequency differences between myelin water and the slow-relaxing water compartment.

4404
Computer 31
Myelin Water Imaging Profiles Along White Matter Tracts
Tobias R Baumeister1, Shannon Kolind2,3,4, Alex MacKay3,4, and Martin McKeown2

1School of Biomedical Engineering, The University of British Columbia, Vancouver, BC, Canada, 2Department of Medicine, Division of Neurology, The University of British Columbia, Vancouver, BC, Canada, 3Department of Radiology, The University of British Columbia, Vancouver, BC, Canada, 4Department of Physics & Astronomy, The University of British Columbia, Vancouver, BC, Canada

Myelin water fraction (MWF) maps are spatially noisy. Here we investigated a possible inherent spatial structure of MWF values along diffusion tensor imaging (DTI)-derived white matter (WM) tracts in 41 healthy subjects. Sixteen major fibre bundles were extracted and MWF was computed in sub-segments along each fibre tract and compared to surrounding voxels. MWF values were more spatially coherent along fibre bundles than elsewhere. The profile along the trajectory of fibre bundles estimated subjects’ age more accurately than tract-averaged MWF. We conclude that the spatial MWF distribution in WM consistently follows a distinct pattern along underlying fibre bundles across subjects.

4405
Computer 32
Single-point macromolecular proton fraction mapping at 7T in healthy and demyelinated mouse brain
Lucas Soustelle1,2, Maria-Cristina Antal2, Paulo Loureiro de Sousa2, and Laura Harsan2

1Aix-Marseille Univ. CRMBM UMR 7339, Marseille, France, 2Université de Strasbourg, CNRS, ICube, FMTS, Strasbourg, France

Assessment of myelin content in the brain is essential for monitoring pathologies such as multiple sclerosis. Quantitative MRI methods including quantitative magnetization transfer imaging (qMTI) have been employed in animal and human studies to assess demyelination processes. Animal studies have reported high correlations between myelin content and the macromolecular proton fraction (MPF), a metric derived from qMTI. The single-point MPF mapping method requires the acquisition of a single MT-weighted image, hence reducing protocol scan duration. In this work, we propose the adaptation of this method at 7T in a study involving healthy and demyelinated mice.

4406
Computer 33
Development of quantitative water content mapping in human brain at high magnetic field
Hidehiro Watanabe1, Nobuhiro Takaya1, and Fumiyuki Mitsumori1

1Center for Environmental Measurement and Analysis, National Institute for Environmental Studies, Tsukuba, Japan

The method for quantitative water content mapping of a human brain at high magnetic field was proposed. We demonstrated that B1- is proportional to B1+ in a uniform area even on a nonuniform image measured at 4.7T. B1-s of the reference phantom and a human brain can be compared by measurable B1+s in uniform areas and water content of human brain can be computed from that comparison. Our method was validated in the experiments of the mixture phantom of H2O and D2O. Quantitative water content maps of human brains were obtained by our method.


4407
Computer 34
Global relaxometry and volumetry of the brain using synthetic MR: possible implications for the neurobiology of human brain ageing in healthy adults
Lu Yu1, Chunmei Li1, Bing Wu2, Jianxun Qu2, Yuwei Jiang1, and Min Chen1

1Beijing Hospital, Beijing, China, 2GE Healthcare, Beijing, China

Synthetic MR is an emerging technique capable of providing quantitative relaxation maps and conventional contrast weighted images simultaneously. This study aims to study the relaxation and volumetric characteristics in the ageing process with synthetic MRI. We found volume is a primary metrics for assessing brain ageing and relaxometry may provide additional quantitative biomarkers and possible implications for studying brain ageing.

4408
Computer 35
Fast multiparametric imaging in the brain using a stationary balanced steady state cartesian approach
Christian Guenthner1, Sebastian Kozerke1, and Mathieu Sarracanie1,2,3

1Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland, 2AMT Lab, Department of Biomedical Engineering, University of Basel, Allschwil, Switzerland, 3MIAC AG, Basel, Switzerland

We propose a multiparametric balanced steady-state 3D Cartesian sequence that exploits model based and pattern matching reconstruction strategies for a series of 20 flip angles and repetition times, allowing for the simultaneous quantification of B0, B1+, T1, T2, and proton density. Time-varying signal patterns at the steady state are reached that allow for the acquisition of unique signal patterns in each image voxel for any acquisition scheme. We show the feasibility of our technique in-vivo in the human brain in 11 minutes, here with Cartesian acquisition and no acceleration strategies.

4409
Computer 36
Improving quantification accuracy in whole brain high spatial resolution 3D kinetic mapping: Development of a novel dual temporal resolution DCE-MRI technique
Ka-Loh Li1, Daniel Lewis2, Alan jackson1, Sha Zhao1, and Xiaoping Zhu1

1Division of Informatics, Imaging and Data Sciences, University of Manchester, Manchester, United Kingdom, 2Manchester Centre for Clinical Neurosciences, Salford Royal NHS foundation trust, Manchester Academic Health Science Centre, United Kingdom

Mapping microvascular parameters from DCE MRI traditionally requires a compromise between temporal resolution, spatial resolution, and volume coverage. This study developed a dual-temporal-resolution-based analysis method which concatenates acquired high temporal (HT) and high spatial (HS) tissue contrast agent concentration curves into a unified HTHS merged volume and then pixel-by-pixel reconstructed the HT first pass concentration curve to a HS resolution before undertaking kinetic analysis. In vivo assessment of this method was undertaken in 12 patients with neurofibromatosis type II, and demonstrated the potential of the new method to provide high spatial resolution kinetic map with HT comparable accuracy and quality.

4410
Computer 37
High-resolution 3D T1 and T2 Mapping in the Brain Using Compressed Sensing and Dictionary Fitting
Emilie Mussard1,2,3, Tom Hilbert1,2,3, Christoph Forman4, Reto Meuli2, Jean-Philippe Thiran3, and Tobias Kober1,2,3

1Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, 2Department of Radiology, University Hospital (CHUV), Lausanne, Switzerland, 3LTS5, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, 4Magnetic Resonance, Siemens Healthcare GmbH, Erlangen, Germany

Quantitative magnetic resonance imaging (qMRI) aims at directly measuring physical tissue properties to be more independent from technical influences. However, parameter mapping is often long and 2D-based. In this work, we propose a protocol for 3D brain T1 and T2 mapping accelerated by compressed sensing. To improve T2 accuracy, we also implemented a T1-informed T2 dictionary fitting technique. Preliminary results showed the ability of the protocol to provide T1 and T2 maps at a 1x1x1.2mm3 resolution in 14:05min as well as the accuracy of the mapping. Establishing a fast 3D protocol will enable generating high-resolution atlases as a next step.

4411
Computer 38
Effects of acquisition and compressed sensing reconstruction parameters on 3D-QALAS multi-parameter quantitation and synthetic imaging of the brain
Ken-Pin Hwang1, Marcel Warntjes2, Naoyuki Takei3, Suchandrima Banerjee3, Drew Mitchell1, R. Jason Stafford1, Linda Chi4, and David Fuentes1

1Department of Imaging Physics, The University of Texas M.D. Anderson Cancer Center, Houston, TX, United States, 2SyntheticMR, Linkoping, Sweden, 3GE Healthcare, Waukesha, WI, United States, 4Department of Radiology, The University of Texas M.D. Anderson Cancer Center, Houston, TX, United States

3D QALAS is a promising new technique that simultaneously maps T1, T2, and PD in a single 3D acquisition. We investigate the robustness of this technique to several acquisition and compressed sensing reconstruction parameters in phantom and brain images. Parameter maps were shown to be robust to B1 through the center portion of the slab, while compressed sensing did not demonstrate any effects on parameters in phantom or cause additional artifacts on parameter maps in human brain. 3D QALAS thus presents an attractive quantification method for therapy planning and tissue volume measurement applications.

4412
Computer 39
Influence of SWI Sequences and QSM Reconstruction Methods on Measured Magnetic Susceptibility in Cerebral Veins
Ronja Berg1, Jakob Meineke2, Andreas Hock3, Claus Zimmer1, and Christine Preibisch1

1Department of Neuroradiology, Technical University of Munich, Munich, Germany, 2Philips Research, Hamburg, Germany, 3Philips Healthcare, Hamburg, Germany

Quantitative Susceptibility Mapping (QSM) has recently been used for assessing the cerebral oxygen metabolism. However, a systematic investigation on the most suitable imaging parameters and reconstruction algorithms for determining the venous susceptibility values is missing. Therefore, we investigated both, the impact of flow compensation and accelerated acquisition as well as different reconstruction methods on measured venous susceptibility. Our results suggest that the choice of reconstruction technique can significantly influence the venous susceptibility values while the investigated imaging parameters did not considerably affect its accuracy. Thus, the applied QSM reconstruction technique has to be considered carefully when quantifying the venous oxygenation.

4413
Computer 40
Improving T2 and B1 parametric estimation in the brain with multi spin-echo MR and fusion bootstrap moves solver (FBMS)
Andreia C. Freitas1, Inês Sousa1, Andreia S. Gaspar1, Rui P.A.G. Teixeira2, Joseph V. Hajnal2, and Rita G. Nunes1,2

1ISR-Lisboa/LARSyS and Department of Bioengineering, Instituto Superior Técnico – Universidade de Lisboa, Lisbon, Portugal, 2Centre for the Developing Brain, King's College London, London, United Kingdom

Multi spin-echo (MSE) sequences have been prescribed for efficient T2 mapping. This can be further improved by matching to pre-computed echo-modulation curves (EMC). Previous use of this method to estimate T2 and B1 resulted in bias in the latter. We investigated the possibility to improve B1 by taking advantage of its spatial smoothness, using a fusion bootstrap moves solver (FBMS). The two methodologies were compared using a numerical phantom and in-vivo brain data. While T2 estimation was accurate and equivalent, B1 accuracy was improved using the FBMS. Future work is required to tune the regularization parameters of the FBMS algorithm.

4414
Computer 41
To Evaluate Effect of SENSE and CSENSE on Quantitative T1 and T2 mapping of Human Brain
Dinil Sasi S1, Anup Singh1,2, Rupsa Bhattacharjee1,3, Ayan Debnath1,4, Snekha Sehrawat1, Rakesh K Gupta5, Indrajit Saha3, and Marc Van Cauteren6

1Indian Institute of Technology Delhi, New Delhi, India, 2AIIMS, New Delhi, India, 3Philips India Limited, Gurugram, India, 4University of Pennsylvania, Philadelphia, PA, United States, 5Fortis Memorial Research Institute, Gurugram, India, 6Philips Health Tech Asia Pacific, Tokyo, Japan

Parallel-imaging and compressed-sensing based approaches are playing crucial role in accelerating MRI data acquisition. Objective of the study was to accelerate the data acquisition of T1, T2 and PD-weighted TSE images and to evaluate the accuracy of T1 and T2 mapping in the human brain. Data was acquired using SENSE parallel-imaging and Compressed-SENSE technique for different factors as well as without any acceleration. T1 and T2 values obtained using data with SENSE (upto factor of 3) and CSENSE (upto factor of 6) were comparable to those acquired without any acceleration. Errors in T1 and T2 increased with increase in acceleration factor.

4415
Computer 42
Four angle method for accurate and rapid clinical high-resolution whole-brain mapping of longitudinal relaxation time and proton density with B1 inhomogeneity correction
Abinand Rejimon1, Luis Cortina1, Richard G. Spencer1, and Mustapha Bouhrara1

1National Institute on Aging (NIA), National Institutes of Health (NIH), Baltimore, MD, United States

Changes in longitudinal relaxation time (T1) and proton density (PD) are sensitive markers of microstructural damage associated with different neurological conditions including myelin degradation, axonal loss, inflammation, and edema. In this study, we propose an accurate and rapid approach to mapping T1 and PD with B1 inhomogeneity correction. This four angle method (FAM) is based on the use of four images acquired with different flip angles and short repetition times using the spoiled-gradient recalled-echo sequence available on all preclinical and clinical MRI machines. The accuracy and ease of implementation of the FAM renders it of great potential for clinical investigations.

4416
Computer 43
Towards in-vivo voxel-wise parcellation of human brain cortex
Shahrzad Moeiniyan Bagheri1, Viktor Vegh1, and David Reutens1

1Centre for Advanced Imaging, The University of Queensland, St Lucia, Australia

The research aims to establish the feasibility of developing an automated method for in vivo voxel-wise parcellation of the human brain cortex. We combined our previously proposed residual analysis Magnetic Resonance Fingerprinting (MRF) approach with supervised classification. We show that extraction of a feature vector from a patch of voxels about a voxel of interest improves prediction accuracy by about 10%, as measured using the Area Under the Curve (AUC) metric. Our approach leads to an increase in the prediction accuracy rate for areas of distinct microstructural heterogeneity, such as the primary motor cortex.

4417
Computer 44
Variable Rates Undersampling Scheme for Fast brain T1ρ mapping
Yuanyuan Liu1, Yanjie Zhu1, Jing Cheng1, Xin Liu1, and Dong Liang1,2

1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, 2Research center for Medical AI, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

T mapping requires several T-weighted images with different spin lock times to obtain the T  maps, resulting in a long scan time.Compressed sensing has shown good performance in fast quantitative T mapping. In this work, we developed a variable acceleration rates undersampling strategy to reduce the scan time. A signal compensation with low-rank plus sparse model was used to reconstruct the T-weighted images. Specifically, a feature descriptor was used to pick up useful features from the residual images. Preliminary results show that the proposed method achieves a 5.76-fold acceleration and obtain more accurate T maps than the existing methods.

4418
Computer 45
Fast and whole-brain T2* mapping using QUTE-EPI at 7T
Seonyeong Shin1, Seong Dae Yun1, and N. Jon Shah1,2,3,4

1Institute of Neuroscience and Medicine 4, INM-4, Forschungszentrum Juelich, Juelich, Germany, 2Institute of Neuroscience and Medicine 11, INM-11, JARA, Forschungszentrum Juelich, Juelich, Germany, 3JARA - BRAIN - Translational Medicine, Aachen, Germany, 4Department of Neurology, RWTH Aachen University, Aachen, Germany

Quantification of T2* relaxation time is of great interest as knowledge of it can be used for clinical diagnosis or optimisation of MR imaging parameters. A typical approach to quantify T2* is to acquire multi-echo data. Although this approach is effective, it still requires a substantial acquisition time for whole-brain coverage. This work aims to employ quantitative echo-planar imaging (QUTE-EPI) at 7T for fast and whole-brain T2* mapping. The performance of QUTE-EPI was directly compared to that of a conventional multi-echo gradient-echo sequence (MEGE). The estimated T2* values were quantitatively analysed for the regions of grey matter (GM) and white matter (WM).

4419
Computer 46
Joint T1 and T2 Mapping with Tiny Dictionaries and Subspace-Constrained Reconstruction
Volkert Roeloffs1, Martin Uecker2,3, and Jens Frahm1,3

1Biomed NMR, Max Planck Institute for Biophysical Chemistry, Goettingen, Germany, 2Institute for Diagnostic and Interventional Radiology, University Medical Center Göttingen, Goettingen, Germany, 3partner site Göttingen, German Centre for Cardiovascular Research (DZHK), Goettingen, Germany

Dictionaries as used in multi-parametric mapping are typically very large in size, take long to compute, and scale exponentially with the number of parameters. Here, we break the bond between dictionary size and representation accuracy by two modifications: First, we approximate the Bloch-response manifold by piece-wise linear functions, and second, we allow the sampling grid to be refined adaptively depending on the precision needed. Phantom and in vivo studies demonstrate efficient multi-parametric mapping with tiny dictionaries and subspace-constrained reconstruction. The presented method preserves accuracy and precision with dictionaries reduced in size by a factor of 10 and beyond.

4420
Computer 47
T1 Mapping at 7T Using a Novel Inversion-Recovery Look-Locker 3D-EPI Sequence
Rüdiger Stirnberg1, Yiming Dong1, Jonas Bause2,3, Philipp Ehses1, and Tony Stöcker1,4

1German Center for Neurodegenerative Diseases (DZNE), Bonn, Germany, 2High-Field MR Center, Max Planck Institute for Biological Cybernetics, Tübingen, Germany, 3Department for Biomedical Magnetic Resonance, University of Tuebingen, Tübingen, Germany, 4Department of Physics and Astronomy, University of Bonn, Bonn, Germany

We propose a novel Inversion-Recovery Look-Locker 3D-EPI sequence for rapid T1 mapping. The inherent SNR benefit of a 3D acquisition, segmentation along both phase encode directions and a turbofactor introduced to reduce the number of required inversions can be traded freely for acquisition speed, SNR, resolution and geometric distortions. Aside from quantitative validation, two high-resolution T1 mapping applications are demonstrated at 7T: whole-brain with minimal distortions, and reduced field-of-view with geometric distortions matched to corresponding fMRI data. The results show high T1 accuracy for several turbofactor and flip angle combinations compared to a single-slice inversion-recovery 2D-EPI reference.

4421
Computer 48
Fast quantitative susceptibility reconstruction via total field inversion with L0 norm approximation
Shuhui Cai1, Li Zhang1, Congbo Cai1, and Zhong Chen1

1Department of Electronic Science, Xiamen University, Xiamen, China

Quantitative susceptibility mapping (QSM) is a meaningful MRI technique owing to its unique relation to actual physical tissue magnetic properties. The reconstruction of QSM is usually decomposed into three sub-problems which are solved independently. Here, we propose a fast reconstruction method named as fast TFI based on total field inversion. It accelerates the total field inversion by using specially selected preconditioner and the advanced solution of weighted L0 regularization. Results from gadolinium phantom and in vivo data verified that the new method has good performance.

4422
Computer 49
Six-direction diffusion tensor MRI using a convolutional neural network
Qiyuan Tian1, Berkin Bilgic1,2, Qiuyun Fan1, Chanon Ngamsombat1, Congyu Liao1, Yuxin Hu3, Thomas Witzel1, Kawin Setsompop1,2, Jonathan R Polimeni1,2, and Susie Y Huang1,2

1Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Charlestown, MA, United States, 2Harvard-MIT Division of Health Sciences and Technology, Massachusetts Institute of Technology, Cambridge, MA, United States, 3Radiological Sciences Laboratory, Department of Radiology, Stanford University, Stanford, CA, United States

Diffusion tensor imaging (DTI) is widely used for clinical neuroimaging and neuroscientific research but has traditionally suffered from relatively length acquisition. Here, we propose a new approach to obtain both scalar and orientational DTI metrics from six diffusion-weighted images with optimal directional encoding. Through the careful choice of diffusion directions, we compute initial tensor results that are then denoised using a convolutional neural network. Our results provide comparable scalar and orientational DTI metric maps to those acquired with 90 directions.

4423
Computer 50
Silent 3D Parameter Mapping using Variable Flip Angle Looping Star
Florian Wiesinger1,2, Nikou L Damestani2, David J Lythgoe2, Emil Ljungberg2, Tobias C Wood 2, Mark R Symms2,3, Fernando Zelaya2, Gareth J Barker2, Steven CR Williams2, and Ana Beatriz Solana1

1GE Healthcare, Munich, Germany, 2Department of Neuroimaging, Institute of Psychiatry, Psychology and Neuroscience, King's College London, London, United Kingdom, 3GE Healthcare, London, United Kingdom

This abstract presents a new method for silent and 3D parameter mapping, including proton density (PD) , T1, T2*, and quantitative susceptibility mapping (QSM), by combining Looping Star 3D silent T2*-weighted imaging with the concept of variable flip angle (VFA) PD and T1 mapping. 


Motion Correction: Brain

Exhibition Hall
Thursday 8:15 - 9:15
 Acquisition, Reconstruction & Analysis

4424
Computer 51
Three-dimensional motion correction in Magnetic Resonance Fingerprinting (MRF)
Jan W. Kurzawski1,2, Matteo Cencini3, Pedro A. Gómez4, Rolf F. Schulte5, Giada Fallo2, Alessandra Retico1, Michela Tosetti2,6, Mauro Costagli2,6, and Guido Buonincontri2

1Italian National Institute of Nuclear Physics, Pisa, Italy, 2IMAGO7 research Foundation, Pisa, Italy, 3Department of Physics, University of Pisa, Pisa, Italy, 4Computer Science department, Technical University of Munich, Munich, Germany, 5GE Healthcare, Munich, Germany, 6IRCCS Stella Maris, Pisa, Italy

Two-dimensional MRF is considered to be less sensitive to in-plane motion than conventional imaging techniques. However, in scanning populations prone to rapid and extensive motion, challenges remain. Here, we suggest a two-step 3D MRF procedure that includes the correction of subject motion during the reconstruction. In the first step, we reconstruct the data in small segments consisting of images with equal contrast and calculate the between-segment motion. In the second step, we perform motion correction and use corrected images for matching with dictionary. This results in higher quality of reconstructed images and better precision of quantitative maps.

4425
Computer 52
Motion-corrected and high-resolution anatomically-assisted (MOCHA) reconstruction of arterial spin labelling MRI
Abolfazl Mehranian1, Andrew J Reader1, and Enrico De Vita1

1Department of Biomedical Engineering, School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom

A MOtion-Corrected and High-resolution Anatomically-assisted (MOCHA) reconstruction framework is proposed for ASL MRI. The method simultaneously accounts and corrects for rigid motion and partial volume effects (PVE), and reduces noise by guided high-resolution anatomical MR images without any need for segmentation. The proposed method was compared with standard methods and a 3D linear regression (3DLR) correction method using realistic simulations and in-vivo data. Results show that MOCHA outperforms 3DLR not only in preservation of structural and local details, including simulated lesions, but also in PVE correction of deep grey matter structures, often subject to segmentation errors in conventional methods.

4426
Computer 53
Deep Learning based motion artifact correction improves the quality of cortical reconstructions
Ben A Duffy1, Lu Zhao1, Arthur Toga1, and Hosung Kim1

1Institute of Neuroimaging and Informatics, University of Southern California, los angeles, CA, United States

Cortical reconstruction is prone to failure without high quality structural imaging data. Here, motion simulation was performed on good quality structural MRI images and used to train a regression convolutional neural network to predict the motion-free images as the output. We show that performing retrospective motion correction using a convolutional neural network is able to significantly reduce the number of cortical surface reconstruction quality control failures.

4427
Computer 54
Improved motion correction of submillimetre 7T fMRI time series with boundary-based registration (BBR)
Pei Huang1, Johan D Carlin1, Richard N Henson1, and Marta M Correia1

1MRC-Cognition and Brian Sciences Unit, University of Cambridge, Cambridge, United Kingdom

Here, we present a novel approach of utilizing Boundary-Based Registration for realigning submillimetre 7T fMRI time series. We collected fMRI data from 6 human participants and processed the data using either standard rigid body realignment using SPM or our BBR realignment method. We compared the two pre-processed datasets with multiple metrics (tSNR, fCNR and percentage of variance explained by the model) and show that realigning using BBR consistently outperforms conventional methods.

4428
Computer 55
A fast approach for simultaneous measurement of head motion and induced magnetic field changes using FID navigators
Tess E Wallace1,2, Onur Afacan1,2, Tobias Kober3,4,5, and Simon K Warfield1,2

1Computational Radiology Laboratory, Boston Children's Hospital, Boston, MA, United States, 2Harvard Medical School, Boston, MA, United States, 3Advanced Clinical Imaging Technology, Siemens Healthineers, Lausanne, Switzerland, 4Department of Radiology, University Hospital (CHUV), Lausanne, Switzerland, 5LTS5, Ecole Polytechnique Federale de Lausanne, Lausanne, Switzerland

Incorrect spatial encoding due to subject motion is a dominant source of artifacts in MRI. Even if changes in head pose are measured and corrected, motion-induced perturbations in the local magnetic field are a further source of image degradation, particularly for imaging at longer echo times and higher field strengths. We propose a fast approach for simultaneously measuring head motion and spatiotemporal B0 changes using FID navigators (FIDnavs) and simulation of the acquisition physics. Rigid-body motion and first-order field coefficients estimated from FIDnavs exhibit a high degree of agreement with ground-truth values in both phantom and volunteer experiments.

4429
Computer 56
Robust retrospective correction of 3D golden-ratio radial MRI using electromagnetic tracking
Tess E Wallace1,2, Simon K Warfield1,2, and Onur Afacan1,2

1Computational Radiology Laboratory, Boston Children's Hospital, Boston, MA, United States, 2Harvard Medical School, Boston, MA, United States

Radial MRI is intrinsically more robust to motion than Cartesian sampling; however, if large rotational motion occurs, the uniform sampling of conventional 3D radial acquisitions is disrupted and is difficult to recover retrospectively. The golden angle ratio has been used to generate a quasi-isotropic distribution of spokes over time in 2D, but is limited to fully correct for motion, which occurs in three dimensions. Extending the flexibility of golden-ratio spoke ordering to 3D radial sampling, combined with rigid-body motion tracking using electromagnetic sensors, enables robust retrospective correction by maintaining relatively uniform sampling, even in the presence of large-amplitude rotational motion.

4430
Computer 57
Correction of Out-of-FOV Motion Artifacts using Convolutional Neural Network Derived Prior Image
Chengyan Wang1, Yuan Wu1, Yucheng Liang2, Danni Yang2, Siwei Zhao1, and Yiping P. Du1

1Institute for Medical Imaging Technology (IMIT), School of Biomedical Engineering, Shanghai Jiao Tong university, Shanghai, China, 2Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States

This study presented a new motion correction algorithm with the incorporation of convolutional neural network (CNN) derived prior image to solve the out-of-FOV motion problem. A modified U-net network was developed by introducing motion parameters into the loss function. We assessed the performance of the proposed CNN-based algorithm on 1113 MPRAGE images with simulated oscillating and sudden motion trajectories. Results show that the proposed algorithm outperforms conventional TV-based algorithm with lower NMSE and higher SSIM. Besides, robust reconstruction was achieved with even 20% data missed due to the out-of-FOV motion.

4431
Computer 58
Non-contact measurement of head movements inside a 7 T Scanner using a 16-channel field camera
Laura Bortolotti1, James Antony Smith1, Penny Gowland1, and Richard William Bowtell1

1SPMIC, University of Nottingham, Nottingham, United Kingdom

The extra-cranial magnetic field changes due to changes in head position have been measured in a 7T scanner using a 16-channel field camera and used to estimate the head movements. A partial least squares regression was used to identify the relationship between field changes and head position data that was simultaneously measured using an optical camera. By applying spherical harmonic spatial filtering to the field measurements it was possible to reduce the unwanted effect of chest movement in respiration, and to then predict head position changes with good accuracy. This provides a step forward towards a non-contact motion monitoring technique.

4432
Computer 59
Simulation of external magnetic field changes due to head motion during 7 Tesla MRI scan
Laura Bortolotti1, James Antony Smith1, Penny Gowland1, and Richard William Bowtell1

1SPMIC, University of Nottingham, Nottingham, United Kingdom

One potential method for monitoring the effects of head movement in the scanner consists of using a fixed array of field probes to measure field changes produced outside the head by small changes in head position and angulation. This method has the advantage of requiring neither attachment of markers or probes to the head, nor modification of the imaging sequence. Here, we use realistic head models to simulate the external field changes produced by typical head movements in a 7T scanner and use the results to explore the relationship between the magnetic field perturbation and changes in head position.

4433
Computer 60
Markerless real-time motion correction for 2D RARE: reducing artefacts in clinical T2 and FLAIR MRI
Robert Frost1,2, F. Işik Karahanoğlu1,2, Camilo Jaimes3, Paul Wighton1, Richard L. Robertson3, P. Ellen Grant3,4, M. Dylan Tisdall5, and André J. W. van der Kouwe1,2

1Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Boston, MA, United States, 2Department of Radiology, Harvard Medical School, Boston, MA, United States, 3Department of Radiology, Boston Children's Hospital, Harvard Medical School, Boston, MA, United States, 4Fetal-Neonatal Neuroimaging and Developmental Science Center, Boston Children's Hospital, Boston, MA, United States, 5Radiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, United States

This study investigates high-frequency prospective motion correction (PMC) using markerless face tracking for artefact reduction in clinical T2 and FLAIR MRI. The FOV pose was corrected before the acquisition of each slice in PMC sequences and five subjects were scanned with 1:16 min T2, 4:55 min T2, and 1:50 min FLAIR protocols. The multi-slice segmented RARE sequences showed high sensitivity to changes in head position but use of PMC scans consistently recovered good image quality with higher image sharpness as measured by the Tenengrad metric.

4434
Computer 61
Robust motion regression of resting-state data using a convolutional neural network model
Zhengshi Yang1, Xiaowei Zhuang1, Karthik Sreenivasan1, Virendra Mishra1, and Dietmar Cordes1,2

1Cleveland Clinic Lou Ruvo Center for Brain Health, Las Vegas, NV, United States, 2University of Colorado, Boulder, CO, United States

The fluctuation introduced by head motion considerably confounds the interpretation of resting-state fMRI data. Specifying motion regressors without taking fMRI data itself into consideration may not be sufficient to model the impact of head motion. We proposed a robust and automated deep neural network (DNN) to derive motion regressors with both fMRI data and estimated realignment parameters considered. The results show that DNN-derived regressors outperform traditional regressors based on several quality control measurements. 

4435
Computer 62
High resolution 3D GRASE BLADE Arterial Spin Labelling sequence: evaluation of the performance with various level of motion: simulations and validation in volunteers and patients
Manjunathan NANJAPPA1,2, Thomas Troalen2, Matthias Günther3,4,5, Magalie Viallon1,6, and Huber Jörn3

1University of Lyon, INSA, CNRS UMR 5520, INSERM U1206, CREATIS, Lyon, France, 2Siemens Healthcare SAS, Saint-Denis, France, 3Fraunhofer MEVIS, Bremen, Germany, 4University Bremen, Bremen, Germany, 5Mediri GmbH, Heidelberg, Germany, 6Radiology Department, University Hospital of Saint Etienne, Saint Etienne, France

In MRI, longitudinal acquisition protocols such as arterial spin labeling are susceptible to patient motion; this work focused on implementing 3D GRASE with BLADE readout trajectory as an alternative to Cartesian readout to increase robustness of sequence with regards to motion. Virtual data simulation and involuntary patient motion data were used to evaluate the performance of this approach with different levels of patient motion. Image reconstruction embedded with self-referenced custom rigid motion correction algorithm was developed and tested on both simulated and patient data. Results confirming superiority of SNR and motion correction capabilities offered by Blade strategy over Cartesian.

4436
Computer 63
A novel, coil-integrated camera for prospective optical motion correction of brain imaging at 7T
Phillip DiGiacomo1, Elizabeth Tong1, Julian Maclaren1, Murat Aksoy1, Roland Bammer1, Brian Rutt1, and Michael Zeineh1

1Stanford University, Stanford, CA, United States

The advancements in signal to noise ratio (SNR), contrast, and resolution enabled by high-field MR systems may visualize more nuanced brain anatomy and pathology. In order to translate these advancements to the discovery and clinical implementation of novel neuroimaging biomarkers, motion artifact resulting from requisite long scan times must be addressed. Here, we demonstrate a novel prospective optical motion tracking and correction system using a camera seamlessly integrated into the 7T Tx/Rx head coil. The integrated camera allows tracking of head motion by visualizing an optical marker on the forehead of human subjects in a 7T MR system.

4437
Computer 64
Prospective motion correction for 2D slice-selective FISP-MRF in the brain using an in-bore camera system
Gregor Körzdörfer1,2, Mario Bacher1,3, Thomas Kluge1, Randall Kroeker1, Dominik Paul1, Josef Pfeuffer1, Bernhard Hensel2, and Mathias Nittka1

1Siemens Healthcare GmbH, Erlangen, Germany, 2Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany, 3CHUV, Centre d'Imagerie BioMédicale, Lausanne, Switzerland

In contrast to motion artifacts in conventional MRI, which can often be identified by visual inspection, the effect can be more subtle in quantitative MRI (qMRI) methods such as Magnetic Resonance Fingerprinting (MRF). Subject motion during qMRI scans can lead to altered parameter maps without affecting their morphologic appearance which limits the user’s possibility to assess the scan quality. One way to mitigate motion artifacts is to track the subject’s movement and prospectively correct for the motion. Here, we present results of applying prospective motion correction using an in-bore camera system for MRF.

4438
Computer 65
Towards motion-robust MRI – Autonomous motion timing and correction during MR scanning using multi-coil data and a deep-learning neural network
Rafi Brada1, Michael Rotman1, Ron Wein1, Sangtae Ahn2, Itzik Malkiel1, and Christopher J. Hardy2

1GE Global Research, Herzliya, Israel, 2GE Global Research, Niskayuna, NY, United States

We propose a method for timing and correcting for rigid-body in-plane patient motion during an MRI scan. The motion is detected using differences between coil-intensity-corrected images from different coils in the receiver array together with the scan-order information. The method allows for the detection and timing of multiple movements during the scan. For each scan where motion was detected, k-space data are divided into different motion states, which are used as input to a deep neural network whose output is a motion-corrected image. The system shows promising results on MR data containing simulated and real motion.

4439
Computer 66
Investigation of the impact of receive field sensitivity on motion corruption in 3D-EPI for fMRI
Nadine N Graedel1, Nadege Corbin1, Yael Balbastre1, Oliver Josephs1, and Martina F Callaghan1

1Wellcome Centre for Human Neuroimaging, UCL Institute of Neurology, London, United Kingdom

High temporal signal-to-noise ratio (tSNR) is crucial in fMRI to maximise functional sensitivity. The use of high-density receiver arrays can greatly improve tSNR and enables parallel imaging, a requirement for imaging with high spatial resolution while maintaining reasonable scan times. The 3D-EPI approach enables through plane acceleration but at the cost of increased motion sensitivity. Here we explore the impact of rapidly varying sensitivity fields on the degradation of tSNR in the presence of motion in the context of 3D-EPI.

4440
Computer 67
Direct comparison of fat navigators and Moiré phase tracking for retrospective brain motion correction at 7T
Frederic Gretsch1, Hendrik Mattern2, Daniel Gallichan3, and Oliver Speck2,4,5,6

1LIFMET, EPFL, Lausanne, Switzerland, 2Department of Biomedical Magnetic Resonance, Institute of Experimental Physics, Otto-von-Guericke-University, Magdeburg, Germany, 3CUBRIC, School of Engineering, Cardiff University, United Kingdom, 4German Center for Neurodegenerative Disease, Magdeburg, Germany, 5Center for Behavioral Brain Sciences, Magdeburg, Germany, 6Leibniz Institute for Neurobiology, Magdeburg, Germany

Retrospective rigid body motion correction based on FatNavs or MPT motion information are directly compared. Both modalities significantly improve image quality of very high resolution anatomical images, but both suffer from drawbacks: rigid marker fixation during long scans for MPT and low temporal resolution for FatNavs. Quantitative analysis confirms these visual observations.

4441
Computer 68
Prospective motion correction for compressed sensing 3D TSE sequence
Patrick Hucker1, Esther Raithel2, Maxim Zaitsev1, and Axel J. Krafft1

1Dept. of Radiology, Medical Physics, Medical Center University of Freiburg, Faculty of Medicine, University of Freiburg, Freiburg, Germany, 2Siemens Healthcare GmbH, Erlangen, Germany

A compressed sensing 3D TSE Sequence prototype (CS-SPACE) was enhanced by prospective motion correction (PMC). For T1-weighted imaging this sequence uses a center-out trajectory along each echo train and sparser sampling with increasing distance from the center. Motion during such echo trains can result in unexpected image artifact behavior. In this work, we investigate whether for a particular echo train structure, a center-out trajectory and compressed sensing PMC can correct for motion artifacts.

4442
Computer 69
Translational Motion Compensation for 3D FSE Parallel Imaging using Autocalibration Signals
Chaoping Zhang1,2, Stefan Klein1,2, Alexandra Cristobal-Huerta2, Juan Antonio Hernandez-Tamames2, and Dirk H.J. Poot1,2

1Department of Medical Informatics, Erasmus MC, Rotterdam, Netherlands, 2Department of Radiology and Nuclear Medicine, Erasmus MC, Rotterdam, Netherlands

Motion during scanning deteriorates MR image quality, especially in 3D fast spin echo (FSE) acquisitions which typically require long acquisition time, even with parallel imaging. Instead of prospective motion compensation which is often difficult to perform, we propose a retrospective translational motion compensation method using autocalibration signals. The proposed method estimates the motion by minimizing the GRAPPA prediction error of the motion corrected signal in the autocalibration signal region. In-vivo experimental results demonstrate the effectiveness of our method.

4443
Computer 70
Video-based head motion assessment for improved quantitative neuroanatomy studies
Heath Pardoe1, Allan George1, Samantha P Martin1, Pablo Velasco2, and Orrin Devinsky1

1Department of Neurology, NYU Langone Medical Center, New York City, NY, United States, 2Center for Brain Imaging, New York University, New York, NY, United States

In-scanner head motion systematically varies with age and diagnosis, and this motion causes bias in morphometric estimates derived from neuroanatomical MRI. There are currently no widely available methods for directly assessing head motion during acquisition of neuroanatomical sequences. In this project we developed a method for measuring head motion via analysis of video obtained from an in-scanner eye tracker. Data obtained from 5 healthy controls demonstrates the feasibility of the technique. The system has minimal set up requirements for subjects or MR technicians, which suggests the technique may be well suited to the young, elderly, or impaired populations in which participant compliance may be a problem.

4444
Computer 71
Super-resolution reconstruction applied to neonatal MRI: multi-orientation vs through-plane slice shift MRI acquisition and segmentation
Nurten Ceren Askin1, Laura Gui Levy1, Joana SaDeAlmeida2, Michel Kocher3, Petra Huppi2, and Francois Lazeyras1

1Radiology, University of Geneva, Geneva, Switzerland, 2Pediatrics, University Hospitals of Geneva, Geneva, Switzerland, 3School of Engineering, EPFL, Lausanne, Switzerland

In this study, the super-resolution (SR) method is used to reconstruct high-resolution MRI volumes from multi-orientation and through-plane shift low-resolution neonatal MRI. Multi-orientation low-resolution images yield higher quality SR results than through-plane shift low-resolution images. SR reconstructed volumes and high-resolution volumes from the scanner are segmented with a morphology-based segmentation algorithm. Segmentation quality is similar between the SR reconstructed volume and the high-resolution volume. Since low-resolution acquisitions are faster, they are less prone to motion artifacts, and thus the reconstructed SR volumes are an alternative to lengthy high-resolution acquisitions. 

4445
Computer 72
Motion correction in Brain MR imaging using a Structure Light based Optical MOtion Tracking system (SLOMO)
Chunyao Wang1, Chen Zhang1, Yu Wang1, Haikun Qi2, Tianqi Huang1, Jin Liu3, Chun Yuan1,3, Hongen Liao1, and Huijun Chen1

1School of Medicine, Tsinghua University, Beijing, China, 2School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, 3Vascular Imaging Laboratory, Department of Radiology, University of Washington, Seattle, WA, United States

Motion artifact is an important challenge in MR imaging. Optical tracking based motion correction technique has been verified effective with the advantages of perfect accuracy, real-time performance and no effect on sequence and scan time. However, most traditional system need an additional Reflective Marker to trace and quantify the motion parameters, which complicated the scan procedure. Recently, our group proposed a markerless optical tracking solution(NORMS) and validated its ability in non-rigid motion detection and correction for carotid artery imaging. In this study, we aim to develop a parallel line Structure Light based Optical Motion Tracking system (SLOMO) to accurately correct rigid motion by acquiring the whole 3D surface. The results demonstrated the feasibility of SLOMO system in motion correction for brain imaging. 

4446
Computer 73
Improvement of Glutamate Chemical Exchange Saturation Transfer (GluCEST) Imaging in a Rat Model of Epileptic Seizure Using Retrospective Motion Correction
Dong-Hoon Lee1, Do-Wan Lee2, Jae-Im kwon3, Chul-Woong Woo3, Sang-Tae Kim3, Jin Seong Lee4, Choong Gon Choi4, Kyung Won Kim4, Jeong Kon Kim4, and Dong-Cheol Woo3,5

1Faculty of Health Sciences and Brain & Mind Centre, The University of Sydney, Sydney, Australia, 2Center for Bioimaging of New Drug Development, and MR Core Laboratory, Asan Institute for Life Sciences, Asan Medical Center, Seoul, Korea, Republic of, 3MR Core Laboratory, Asan Institute for Life Sciences, Asan Medical Center, Seoul, Korea, Republic of, 4Department of Radiology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Korea, Republic of, 5Department of Convergence Medicine, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Korea, Republic of

GluCEST is a novel molecular MR imaging technique to detect glutamate in the brain parenchyma by measuring the exchange of glutamate amine protons with bulk water. However, a disadvantage of CEST imaging is the relatively long scan time required to collect the data while varying the resonance frequency around the water. In this abstract, we describe the application of a retrospective motion correction approach using a gradient-based motion correction (GradMC) algorithm to CEST data for investigating the feasibility of motion correction, using an epileptic seizure rat model with head motion. Our results clearly show that the GradMC can be used in CEST imaging to efficiently correct for motion.

4447
Computer 74
Deep learning motion compensation for Cartesian and spiral trajectories
Quan Dou1, Xue Feng1, Zhixing Wang1, Daniel Weller2, and Craig Meyer1

1Biomedical Engineering, University of Virginia, Charlottesville, VA, United States, 2Electrical and Computer Engineering, University of Virginia, Charlottesville, VA, United States

Movement of the subject during MRI acquisition causes image quality degradation. In this study we adopted a deep CNN to correct motion-corrupted brain images. To get paired training datasets, synthetic motion artifacts were added by simulating k-space data along different sampling trajectories. Quantitative evaluation showed that the CNN significantly improved the image quality. The spiral trajectory performed better than the Cartesian trajectory both before and after the motion deblurring. A network trained with an L1 loss function achieved better RMSE and SSIM than one trained with an L2 loss function after convergence. Overall, deep learning yields rapid and flexible motion compensation.

4448
Computer 75
Deep learning based motion estimation from highly under-sampled EPI volumetric navigators
Mykhailo Hasiuk1,2, Kamlesh Pawar1,3, Shenjun Zhong1, Richard McIntyre1, Zhaolin Chen1,4, and Gary Egan1,3

1Monash Biomedical Imaging, Monash University, Melbourne, Australia, 2Department of Material Sciences and New Technologies, Vasyl Stefanyk Precarpathian National University, Ivano-Frankivsk, Ukraine, 3School of Psychological Sciences, Monash University, Melbourne, Australia, 4Department of Electrical and Computer System Engineering, Monash University, Melbourne, Australia

Dynamic EPI volumetric navigators are widely used to track head motion in MRI, and accurate motion estimation requires EPI volumes to be inserted in every several seconds or even less. However, the use of dynamic EPI volumes to track motion significantly degrades the overall data acquisition efficiency. To address this issue, in this work we introduce a deep learning based motion estimation method from highly under-sampled (i.e. acceleration factor of 16) EPI volumetric navigators. The method directly estimates motion parameters from the under-sampled data, and does not require reconstruction of images.


Robust & Reproducible Quantitation

Exhibition Hall
Thursday 8:15 - 9:15
 Acquisition, Reconstruction & Analysis

4449
Computer 76
Fast Multi-Parametric Mapping Competition: MR Fingerprinting vs. Triple-Echo Steady State
Christian Guenthner1, Thomas Amthor2, Sebastian Kozerke1, and Mariya Doneva2

1Institute for Biomedical Engineering, ETH Zurich, Zurich, Switzerland, 2Philips Research Europe, Hamburg, Germany

Magnetic Resonance Fingerprinting (MRF) and triple-echo steady-state (TESS) are two sequences that both allow for the simultaneous quantification of T1 and T2. While MRF relies on the transient response of tissue and noise-like under-sampling artifacts, TESS acquires the two lowest order SSFP-FIDs and the lowest order SSFP-Echo in the steady-state of a rapid, spoiled SSFP sequence.

In this work, we compare the performance of the two sequences in a phantom study, where imaging parameters and total acquisition duration between the two scan techniques were matched. In addition, a slice-profile correction for TESS is proposed and included in the comparsion.


4450
Computer 77
Experimental Validation of Augmented Fractional MR Fingerprinting
Lixian Zou1,2, Haifeng Wang1, Huihui Ye3, Shi Su1, Xin Liu1, and Dong Liang1,4

1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, 2Shenzhen College of Advanced Technology, University of Chinese Academy of Sciences, Shenzhen, China, 3State Key Laboratory of Modern Optical Instrumentation, College of Optical Science and Engineering, Zhejiang University, Zhejiang, China, 4Research Center for Medical AI, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

Magnetic resonance fingerprinting is a time-efficient acquisition and reconstruction framework to provide simultaneous measurements of multiple parameters including the T1 and T2 maps. The accuracy of the mapping dictionary of MRF is very important for its clinical applications. In this work, we validated the dictionary performance of the augmented fractional order Bloch equations on MRF in the experimental phantom study. Representative results of experimental phantom demonstrate that the utilization of the augmented fractional model is able to improve the accuracy of the T1 and T2 values.

4451
Computer 78
The impact of shorter acquisition time in MRF: Long term repeatability and reproducibility study on ISMRM/NIST phantom and volunteers.
Yutaka Kato1, Kazushige Ichikawa1, Toshiaki Taoka2, Hirokazu Kawaguchi3, Katsutoshi Murata3, Katsuya Maruyama3, Gregor Koerzdoerfer4,5, Josef Pfeuffer4, Mathias Nittka4, and Shinji Naganawa2

1Department of Radiological Technology, Nagoya University Hospital, Nagoya, Japan, 2Department of Radiology, Nagoya University Hospital, Nagoya, Japan, 3MR Research & Collaboration, SIEMENS Healthcare K.K., Tokyo, Japan, 4Magnetic Resonance, Siemens Healthcare GmbH, Erlangen, Germany, 5Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany

This study focused on the stability of MRF in a phantom and volunteers, and explored the feasibility of MRF with a shorter acquisition time. Phantom scans on 40 days and volunteer scans on 5 days over 3 months showed comparable repeatability and reproducibility of T1 and T2 values between MRF with acquisition times of 41 sec and 20 sec. Shorter acquisition time has the potential to expand the clinical usage of MRF.

4452
Computer 79
Cross-system reliability for rapid quantitative MRI
Catharina Petersen1, Peter Johansson1, and Marcel Warntjes1,2

1SyntheticMR AB, Linköping, Sweden, 2Center for Medical Image Science and Visualization, Linköping, Sweden

Absolute quantification of R1 and R2 relaxation rates and proton density PD has been gaining considerable attention in recent years. It is of utmost importance that these measurements entirely reflect patient properties and no influence is detectable on which specific MRI scanner system the quantitative maps were obtained. The SyMRI software was verified on Philips, GE and Siemens scanners at both 1.5T and 3T showing cross-system reliability.

4453
Computer 80
Repeatability of T2 Relaxation Measurements over a Four-Year Period
Xing Wang1,2, Cheryl R McCreary2,3, Marina Salluzzi3,4, and Richard Frayne1,2

1Medical Sciences, Radiology and Clinical Neurosciences, Hotchkiss Brain Institute, University of Calgary, Calgary, AB, Canada, 2Seaman Family MR Centre, Foothills Medical Centre, Alberta Health Services, Calgary, AB, Canada, 3Radiology and Clinical Neurosciences, Hotchkiss Brain Institute, University of Calgary, Calgary, AB, Canada, 4Calgary Image Processing and Analysis Centre (CIPAC), Seaman Family MR Centre, Calgary, AB, Canada

The reliability of a T2 relaxation quantification technique was assessed by repeatedly scanning four subjects (total of 12 scans at 4 time points over 4 years). Both total, biological and scanner variability were assessed across the whole brain and in the frontal, occipital, parietal temporal lobes. Total variability (coefficient-of-variation CoV < 10.3%) was dominated by biological variation (CoV < 10.3%). Scanner variability was low (CoV < 1.6%) despite scanner software and hardware upgrades during this interval. These results suggest that quantitative T2 estimates are reproducible over 4 years and robust to scanner upgrades.

4454
Computer 81
The traveling heads 2.0: Reproducibility of quantitative imaging methods at 7 Tesla
Maximilian N Voelker1,2, Oliver Kraff1, Steffen Goerke3, Frederik B Laun4, Kerrin J Pine5, Philipp Ehses6, Moritz Zaiss7, Andrzej Liebert4, Sina Straub3, Korbinian Eckstein8, Simon Robinson8, Armin M Nagel3,4, Oliver Speck9,10, Mark E Ladd1,3, and Harald H Quick1,2

1Erwin L. Hahn Institute for Magnetic Resonance Imaging, University Duisburg-Essen, Essen, Germany, 2High-Field and Hybrid MR Imaging, University Hospital Essen, Essen, Germany, 3Division of Medical Physics in Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany, 4Institute of Radiology, University Hospital Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany, 5Department of Neurophysics, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany, 6German Center for Neurodegenerative Diseases (DZNE), Bonn, Germany, 7Max Planck Institute for Biological Cybernetics, Tübingen, Germany, 8High Field MR Center, Department for Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Vienna, Austria, 9Otto-von-Guericke-University Magdeburg, Magdeburg, Germany, 10Leibniz Institute for Neurobiology, Magdeburg, Germany

The “traveling heads” is a study to assess the comparability and reproducibility of multicenter human brain imaging at 7T. In previous experiments, we compared typical UHF sequences for structural brain imaging. In this study, we focus on the reproducibility of quantitative imaging and compare methods for volumetry, relaxometry, QSM and CEST between different sites. In addition, three generations of 7T MR systems are compared, i.e. the older installed base consisting of passively and actively shielded magnets of the first and second generation, respectively, as well as the most recent generation which has been approved as a medical device.

4455
Computer 82
Knee T2 relaxometry using quantitative DESS: reproducibility across imaging vendors
Quin Lu1, Brian A Hargreaves2,3,4, Dave Hitt1, and Akshay S Chaudhari2,4

1Philips Healthcare North America, Gainesville, FL, United States, 2Radiology, Stanford University, Stanford, CA, United States, 3Electrical Engineering, Stanford University, Stanford, CA, United States, 4Bioengineering, Stanford University, Stanford, CA, United States

T2 is a promising MR-based biomarkers for early diagnosis of osteoarthritis (OA). Studies have shown that quantitative DESS (qDESS) is capable of performing simultaneous knee morphometry and T2 relaxometry. In this study, we investigate the cross-vendor reproducibility of knee T2 relaxometry using qDESS. By comparing measured cartilage and meniscus T2 values in volunteers scanned on both Philips 3T and GE 3T scanners, we show that qDESS has good intra-vendor scan-rescan repeatability (CCC = 99.2% and 98.8% ) and  cross-vendor reproducibility (CCC=96.3%). With continued effort, we hope to show that qDESS T2 relaxometry can serve as a reliable clinical biomarker for early OA diagnosis.     

4456
Computer 83
Inter-site Reproducibility of Cardiac Magnetic Resonance Fingerprinting T1 and T2 Quantification in the ISMRM/NIST MRI System Phantom and Human Heart
Yuchi Liu1, Luuk H.G.A. Hopman2, Jesse Hamilton1, Elizabeth Hillier2, Matthias Friedrich2, and Nicole Seiberlich1,3

1Department of Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 2Research Institute of the McGill University Health Center, Montreal, QC, Canada, 3Department of Radiology, University Hospitals Cleveland Medical Center, Cleveland, OH, United States

Cardiac Magnetic Resonance Fingerprinting (cMRF) is a novel technique for simultaneous T1 and T2 quantification in the myocardium. Because cMRF has the potential to take heart rate variations and any variable system properties into account, it is hypothesized that cMRF will enable more reproducible measurements of T1 and T2. The purpose of this study is to evaluate the inter-site reproducibility of cMRF. Excellent agreement of cMRF measurements between two sites (University Hospitals Cleveland Medical Center, Cleveland, US and McGill University Health Center, Montreal, CA) was achieved in the ISMRM/NIST phantom and in the hearts of healthy subjects.

4457
Computer 84
Exploring the sensitivity of Magnetic Resonance Fingerprinting to k-space trajectory uncertainties
Alessandro Palombit1,2, Alessandra Bertoldo1,2, Zidan Yu3,4,5, Riccardo Lattanzi3,4,5, and Martijn Anton Cloos3,4,5

1Information Engineering, University of Padova, Padova, Italy, 2Padova Neuroscience Center, Padova, Italy, 3Center for Advanced Imaging Innovation and Research (CAI2R), New York, NY, United States, 4Bernard and Irene Schwartz Center for Biomedical Imaging, New York, NY, United States, 5New York University School of Medicine, The Sackler Institute of Graduate Biomedical Sciences, New York, NY, United States

In this work, we experimentally explore the sensitivity of Magnetic Resonance Fingerprinting (MRF) to k-space trajectory uncertainties typically encountered in non-cartesian imaging. We demonstrate that T1 and T2* quantification can be affected by minor gradient delays observed in stack-of-stars 3D MRF implementations, particularly resulting in severely disrupted T2* measures. As a first approximation, we modeled these imperfections as constant readout sampling shifts of a few integer k-space steps along every trajectory direction. We show that by simply shifting back the nominal sampling locations before the reconstruction can restore reliable MRF parametric estimates.

4458
Computer 85
Three-dimensional, fast parameter mapping at 7.0T with SSFP MR Fingerprinting: comparison of radial and spiral projections k-space trajectories
Guido Buonincontri1, Pedro A Gómez2, Matteo Cencini1,3, Mauro Costagli1, Graziella Donatelli1, Rolf F Schulte4, and Michela Tosetti1

1Laboratory of Medical Physics and Magnetic Resonance - IRCCS Stella Maris Foundation and IMAGO7 Foundation, Pisa, Italy, 22Department of Computer Science, Technische Universitat Munchen, Germany;, Munich, Germany, 3University of Pisa, Pisa, Italy, 4GE Healthcare, Munich, Germany

When using ultra-high field MRI scanners (UHF, B0>= 7T), quantitative imaging is challenging due to B0 and B1+ non-uniformities. Magnetic resonance fingerprinting (MRF) represents a great opportunity for quantitative imaging at UHF as it can estimate these effects at the same time of the parameters of interest. Here, we compare two novel 3D SSFP MRF approaches, one based on a three-dimensional spiral projection acquisition and one using a radial acquisition in vivo at 7.0T. We estimate M0, T1, T2 and B1+ simultaneously at high resolution (1mm isotropic) within 6.5 minutes acquisition time.

4459
Computer 86
T1 mapping with golden-angle radial sampling: A comparison of direct and indirect reconstruction
Nikolaos Kallistis1, Ian Rowe1, and Steven Sourbron2

1Leeds Institute of Biomedical and Clinical Sciences, University of Leeds, Leeds, United Kingdom, 2Leeds Institute of Cardiovascular and Metabolic Medicine, University of Leeds, Leeds, United Kingdom

The purpose of the study is to compare a direct model-based reconstruction with an indirect compress sensing reconstruction for the estimation of T1-map, from simulated radial sampled datasets. Comparisons are performed for the binning strategy that is optimal in each case as measured by T1-errors.

The direct reconstruction solves the nonlinear-least-squares optimization problem with a gradient-based L-BFGS algorithm without regularization, while for the indirect method the images are reconstructed using the iGRASP technique.

The accuracy for both methods is similar, however the computational time of the model-based reconstruction is a limiting factor for clinical applications.


4460
Computer 87
Information Quantification of Subsequent Acquisitions for Minimizing Synthetic MRI Reconstruction Uncertainty
Drew Mitchell1, David Fuentes1, Jason Stafford1, James Bankson1, and Ken-Pin Hwang1

1Department of Imaging Physics, The University of Texas M.D. Anderson Cancer Center, Houston, TX, United States

A mutual information-based mathematical framework is developed to quantify the information content of various acquisition parameters and subsampling approaches. A recursive conditional formulation quantifies information content given previous acquisitions. This framework is applied to 3D QALAS. Mutual information between reconstructed M0, T1, and T2 uncertainty and measurement noise is calculated for an in silico phantom and the results applied to measurements on a System Standard Model 130 phantom. Reconstructions from these measurements demonstrate the potential use of information theory in guiding pulse sequence design to maximize reconstruction quality.

4461
Computer 88
The in vivo impact of diffusion spoiling on the estimate of T1 using spoiled gradient echoes with variable flip angles
Nadège Corbin1, Shaihan J Malik2, and Martina F Callaghan1

1Wellcome Centre for Human Neuroimaging, UCL Institute of Neurology, London, United Kingdom, 2School of Biomedical Engineering & Imaging Sciences, King’s College London, London, United Kingdom

Incomplete spoiling of the transverse MRI signal causes errors in the T1 time estimated from variable flip angle measurements acquired with spoiled gradient-echo images.  Diffusion spoiling is thought to lessen these effects. However, these conclusions are based on phantom experiments, using very long T2 times, or from in vivo simulation using infeasibly strong diffusion spoiling. Here we perform simulation and in vivo experiments to characterise the impact of diffusion spoiling in the short T2, low spoiling regime. We show that even under these conditions, diffusion spoiling reduces the dependence of the estimated T1 on both the phase-increment and the transmit field.

4462
Computer 89
Robust quantitative T1rho imaging in the presence of B1 RF and B0 field inhomogeneities
Huimin Zhang1, Baiyan Jiang1, Queenie Chan2, and Weitian Chen1

1Department of Imaging and Interventional Radiology, The Chinese University of Hong Kong, Hong Kong, Hong Kong, 2Philips Healthcare, Hong Kong, China

T1rho is a valuable biomarker to probe macromolecular environment of tissue. However, T1rho imaging suffers from B1 RF and B0 field inhomogeneities. In this work, we present an approach to address this problem. The performance of our proposed method was demonstrated by simulations, phantom and in vivo experiments. 

4463
Computer 90
MR imaging of a 3D-printed bioreactor with a dedicated radiofrequency coil for cellular level validation of quantitative MR metrics
Megan E Poorman1,2, Slavka Carnicka1, Jeanne E Barthold3, Michele N Martin1, Karl F Stupic1, Corey P Neu3, and Kathryn E Keenan1

1National Institute of Standards and Technology, Boulder, CO, United States, 2Department of Physics, University of Colorado Boulder, Boulder, CO, United States, 3Department of Mechanical Engineering, University of Colorado Boulder, Boulder, CO, United States

Quantitative MRI methods have the potential to push the clinical standard of care towards quantitative diagnosis. However, controlled systems are needed to study the effects of underlying cellular properties on the MRI signal to validate quantitative MRI measures. To meet this need, our group previously developed an MR-compatible bioreactor to monitor cell behavior using MRI validated with optical microscopy. The present work develops a dedicated RF coil for improved MR imaging of the bioreactor and uses it to explore the effects of cell culture on T1 and T2 in our system.  

4464
Computer 91
Assessment and optimisation of bias field correction using N4ITK for PD mapping
Sara Lorio1 and David W Carmichael2

1UCL, London, United Kingdom, 2KCL, London, United Kingdom

Proton density (PD) maps measure the amount of free water molecules in the tissue and can be used in a range of neurological disorders. We previously developed a new approach for PD mapping based on a multi-contrast acquisition protocol, and a data-driven estimation method for inhomogeneity correction and map scaling. Here we evaluate the robustness of the inhomogeneity correction method and its effect on the PD value estimation using data acquired with different receiver coils. This allowed us to assess the impact of the spatial variability of the receiver coil profile on the PD map.

4465
Computer 92
Performance comparison of channel combination methods for multi-echo chemical shift-encoded MRI
Nathan Tibbitts Roberts1,2, Timothy J Colgan1, Kang Wang3, and Diego Hernando1,4

1Radiology, University of Wisconsin - Madison, Madison, WI, United States, 2Electrical and Computer Engineering, University of Wisconsin - Madison, Madison, WI, United States, 3Global Applications and Workflow, GE Healthcare, Madison, WI, United States, 4Medical Physics, University of Wisconsin - Madison, Madison, WI, United States

Many coil combination methods have been developed, including methods that require pre-calibrated coil sensitivity maps, as well as methods that do not require additional sensitivity maps. Several of these methods have been adapted to CSE-MRI, where accurate signal combination is particularly critical as it needs to preserve consistent phase and magnitude information across echoes; however, their relative performance remains unknown.  Therefore, the purpose of this work is to compare theoretically, in simulation, and experimentally the bias and noise performance of quantitative parameter maps resulting from five commonly used coil multi-echo coil combination techniques.

4466
Computer 93
Correlating the measured fast/slow decay components of tissue sodium to the intra-/extra-cellular sodium concentrations
Jinhu Xiong1, William R Kearney1, Mathews Jacobs1, Rolf Schulte2, Baolian Yang2, and Vincent Magnotta1

1University of Iowa, Iowa City, IA, United States, 2GE Healthcare, Milwaukee, WI, United States

We have developed two models (anisotropic-anisotropic (AAS) vs. anisotropic-isotropic (AIS) models) to correlate the measured fast/slow decay components of tissue sodium to the intra-/extra-cellular sodium concentrations.  The models were evaluated based on theoretical and experimental results.   Our results indicate that AAS model fits experimental data much better than AIS model does.

4467
Computer 94
In-vivo cardiac DTI using motion compensated optimized diffusion encoding (MODE): A reproducibility study
Prateek Kalra1, Waqas Majeed1, Mohammad R. Maddah1, Xiaokui Mo2, Richard D. White1, and Arunark Kolipaka1

1Radiology, Ohio State University Wexner Medical Center, Columbus, OH, United States, 2Center for Biostatistics, Ohio State University Wexner Medical Center, Columbus, OH, United States

Diffusion-weighted imaging (DWI) is used to identify heterogeneous infarcted region by calculating ADC(apparent diffusion coefficient) and FA(fractional anisotropy). However, performing DWI in heart is very challenging because of heart motion. Earlier method used convex optimized diffusion encoding (CODE) to optimize diffusion encoding gradients (DEG) waveform. However, due to limitations of CODE waveforms, earlier we proposed motion compensated diffusion encoding (MODE) to achieve higher b-value for a given DEG duration. The aim of this study is to validate and assess the reproducibility of MODE technique in computing ADC and FA maps in healthy subjects. Preliminary results demonstrated good reproducibility using MODE.

4468
Computer 95
Repeatability and Reproducibility of brain volume measurements with SPM and Freesurfer  and their impact on subtle between-group differences
Letizia Palumbo1, Paolo Bosco1, Elisa Ferrari2, Piernicola Oliva3, Giovanna Spera1, and Alessandra Retico1

1National Institute for Nuclear Physics (INFN), Pisa, Italy, 2Scuola Normale Superiore, Pisa, Italy, 3University of Sassary and INFN Cagliari Division, Sassari, Italy

The main aim of the study is to investigate whether the adoption of a processing method has a relevant influence on the results of a neuroimaging research. We evaluated the intra-method repeatability and the inter-method reproducibility of two widely-used automatic segmentation methods for brain MRI: FreeSurfer (FS) and Statistical Parametric Mapping (SPM) software packages. We segmented the gray matter, the white matter and the subcortical structures in test-retest MRI data of healthy volunteers from two publicly available datasets. High intra-method repeatability was found for both SPM and FS, but SPM was more consistent than FS in measuring ROIs volumes.

4469
Computer 96
The Effect of Membrane Lipids on qMT Exchange Constants
Oshrat Shtangel1 and Aviv Mezer1

1The Hebrew University, Jerusalem, Israel

Quantitative Magnetization Transfer(qMT) is a proposed method for deeper characterization of brain tissue. Yet, a connection between qMT parameters and the components of cellular tissue is required. Myelin is composed of various types of lipids, which their amount and composition are changed between brain areas, disease states and across the lifespan. In this work, we formulated liposomes to model the environment of abounded lipids in the human brain and systematically estimated their effect on qMT parameters. We found qMT technique useful to identify differences between lipids. This result can pave the way to future research the molecular environments of human tissue in-vivo.

4470
Computer 97
Repeatability of radiomics features in double baseline MR imaging of glioblastoma
Katharina V Hoebel1,2, Andrew L Beers1, James M Brown1, Ken Chang1,2, Jay B Patel1,2, Marco C Pinho1, Bruce R Rosen1, Tracy T Batchelor3, Elizabeth R Gerstner3, and Jayashree Kalpathy-Cramer1

1Athinoula A. Martinos Center for Biomedical Imaging, Charlestown, MA, United States, 2Harvard-MIT Division of Health Sciences and Technology, Cambridge, MA, United States, 3Stephen E. and Catherine Pappas Center for Neuro-Oncology, Massachusetts General Hospital, Boston, MA, United States

Extraction of radiomic features has to be repeatable in order to be clinically useful. We investigated the repeatability of radiomic feature extraction on a unique dataset consisting of a double baseline MRI scans in 48 patients diagnosed with glioblastoma. Size and shape features which are mostly governed by tumor segmentation showed on average higher repeatability than intensity and texture-based features which are more dependent on image acquisition and preprocessing. More research on the influence of image acquisition and preprocessing on the repeatability and reliability of radiomic features has to be undertaken to make radiomics a safe image-analysis tool.

4471
Computer 98
Influence of image processing on the robustness of radiomic features derived from magnetic resonance imaging - a phantom study
Barbara D. Wichtmann1, Ulrike I. Attenberger1, Felix N. Harder1, Stefan O. Schönberg1, David Maintz2, Kilian Weiss3, Daniel Pinto dos Santos2, and Bettina Baeßler1,2

1Institute of Clinical Radiology and Nuclear Medicine, University Medical Center Mannheim, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany, 2Department of Radiology, University Hospital of Cologne, Cologne, Germany, 3Philips Healthcare Germany, Hamburg, Germany

The emerging field of radiomics aims at analyzing quantitative features derived from medical images to characterize tissue and support clinical decision-making. However, an extensive knowledge about the robustness and reproducibility of radiomic features has to precede translation into routine clinical practice. Very little is known regarding the robustness of radiomic features derived from magnetic resonance imaging. We want to assess the influence of image processing parameters on the robustness of radiomic features. Our results showed that variation of image processing parameters has considerable impact on the robustness as well as on absolute values of radiomic features extracted from MRI data.

4472
Computer 99
Repeatability of radiomic features for prostate cancer diffusion weighted imaging obtained using b-values up to 2000 s/mm2
Harri Merisaari1, Rakesh Shiradkar2, Jussi Toivonen1, Amogh Hiremath2, Mohammadhadi Khorrami2, Ileana Montoya Perez1, Tapio Pahikkala1, Pekka Taimen1, Janne Verho1, Peter J Boström3, Hannu Aronen1, Anant Madabhushi2, and Ivan Jambor1,4

1University of Turku, Turku, Finland, 2Case Western Reserve University, Cleveland, OH, United States, 3Turku University Hospital, Turku, Finland, 4Icahn School of Medicine at Mount Sinai, New York, NY, United States

We evaluated repeatability and diagnostic performance of commonly used radiomic features for prostate cancer (PCa) DWI obtained using b values up to 2000 s/mm2.  Forty-eight men with diagnosed PCa under two repeated 3T MRI examinations performed on the same day. Whole mounts prostatectomy sections were manually matched with in-vivo MRI data. Fourteen of the evaluated 575 features demonstrated high repeatability with ICC(3,1)>0.9 and AUC(Gleason score 3+3 vs >3+3 PCa)>0.6. Many of the conventional radiomics feature demonstrate high AUC but low repeatability (low ICC(3,1) values)stressing the fact that high classification potential using single acquisition does not necessarily mean good overall performance.


Motion Correction: Non-Brain

Exhibition Hall
Thursday 8:15 - 9:15
 Acquisition, Reconstruction & Analysis

4473
Computer 101
Comparison of free-breathing motion-resolved radial imaging with standard breath-hold imaging on liver MRI: a feasibility study
Kate Anne Harrington1, Feng Li2, Arifa Chowdhury1, Kang Wang 3, Ty Cashen3, Ali Ersoz 3, Maggie Fung3, Ersin Bayram 3, Kinh Gian Do1, and Ricardo Otazo1,2

1Department of Radiology, Memorial Sloan Kettering Cancer Center, New York, NY, United States, 2Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, United States, 3GE Healthcare, Waukesha, WI, United States

Respiratory motion remains a major challenge in clinical abdominal MRI. Recent technical advances using continuous radial imaging during free-breathing and motion-resolved compressed sensing-based image reconstruction have demonstrated improvements in motion robustness over conventional motion-gated or motion-corrected techniques, but they were not validated for liver imaging. This work implemented extra-dimensional (XD) reconstruction for free-breathing RadialLAVA acquisitions and compared it against conventional breath-held CartesianLAVA. We demonstrate that the XD technique matches that, and in some instances, is superior to that of standard breath-hold technique in terms of overall image quality in the evaluation of post-contrast liver images. 

4474
Computer 102
Respiratory Motion Signals Extracted from 3D Image-Based Navigation and from PCA of SI-Projections: Initial Findings in Whole-Heart Imaging using a Free-Running Framework
John Heerfordt1,2, Nemanja Masala1, Lorenzo Di Sopra1, Christopher W. Roy1, Bastien Milani1, Jérôme Yerly1,3, Jessica A.M. Bastiaansen1, Davide Piccini1,2,4, and Matthias Stuber1,3

1Department of Radiology, University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland, 2Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, 3Center for Biomedical Imaging (CIBM), Lausanne, Switzerland, 4LTS5, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland

In continuously acquired whole-heart coronary MRA, it is not fully understood how unitless respiratory self-gating signals relate to actual respiratory displacement and drift. Therefore, self-gating signals extracted from principal component analyses of 1D projections oriented in the superior-inferior (SI) direction were compared to image-based navigators. Whole-heart data from continuous uninterrupted 3D radial bSSFP acquisitions were used to reconstruct time series of 3D sub-images with a temporal resolution of 0.6 seconds. Preliminary findings suggest that the SI-directed motion obtained from these sub-images is better described by respiratory self-gating signals created from three principal components rather than from one principal component alone. 

4475
Computer 103
DCE-Abdominal MR Image Registration using Convolutional Neural Networks
Zachary Miller1 and Kevin Johnson1

1University of Wisconsin-Madison, Madison, WI, United States

Convolutional neural networks (CNNs) have had incredible success solving image segmentation problems. We explore whether CNNs could have a similar level of success on difficult image registration problems. To this end, we developed a modified U-net to remove respiratory motion, but preserve contrast changes in abdominal free breathing dynamic contrast enhanced (DCE)-MRI. We then compared this network to a state of the art iterative registration algorithm. We demonstrate that our modified U-net outperforms iterative methods both in terms of registration quality and speed (600 registrations in <1 sec vs. Elastix in 2 hours)

4476
Computer 104
Data Consistency Driven Correction of B0-Fluctuations in 2D and 3D Gradient-Echo MRI of the Spine
Jakob Meineke1 and Tim Nielsen1

1Philips Research, Hamburg, Germany

We demonstrate the data-consistency driven determination and correction of B0-fluctuations induced by respiratory motion in 2D and 3D gradient-echo images of the cervical spine. By promoting data-consistency in the multi-channel raw data, it is possible to estimate the instantaneous off-resonance. Furthermore, we demonstrate a marked improvement in image quality by correcting the k-space data using the measured B0-fluctuations.

4477
Computer 105
Unraveling the effect of spatial resolution and scan acceleration on 3D image-based navigators for respiratory motion tracking
Srivathsan P. Koundinyan1, Mario O. Malavé1, and Dwight G. Nishimura1

1Electrical Engineering, Stanford University, Stanford, CA, United States

Beat-to-beat 3D image-based navigators (3D iNAVs) enable nonrigid respiratory motion tracking of the heart. In this work, we study the accuracy of motion information extracted from 3D iNAVs with different choices of two parameters: spatial resolution and scan acceleration factor. We demonstrate that high spatial resolution coupled with aggressive scan acceleration results in residual blurring and aliasing following iterative reconstruction, which corrupts the derived motion estimates. Through simulations, we identify the optimal combination of spatial resolution and scan acceleration for acquiring 3D iNAVs. In vivo studies presenting sharp motion correction outcomes demonstrate a capability for monitoring motion with high fidelity.

4478
Computer 106
Self-navigation Liver Respiratory Motion Correction Based on Deep Learning
Yu Wang1, Haikun Qi2, Guanhua Wang1, Yuze Li1, and Huijun Chen1

1Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, 2School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom

Correction of respiratory motion with 100% acquisition efficiency is of great significance for clinical abdominal imaging. In this study, we propose a novel self-navigation liver respiratory motion correction method for 3D radial sampling. This new approach is based on the fact that radial acquisition enables oversampled k-space center to extract motion-state signal and neural network can be used for data dimensionality reduction. Both regular and irregular hepatic breathing experiments were conducted and the proposed method has shown similar reconstruction image quality with bellow.

4479
Computer 107
Free breathing & Ungated Multi-Slice cardiac cine MRI using spiral-SToRM
Abdul Haseeb Ahmed1, Sunrita Poddar1, Stanley Kruger2, Prashant Nagpal3, Rolf Schulte4, and Mathews Jacob1

1Electrical Engineering, University of Iowa, Iowa city, IA, United States, 2Biomedical Engineering, University of Iowa, Iowa city, IA, United States, 3Radiology, University of Iowa Hospitals and Clinics, Iowa city, IA, United States, 4Healthcare, GE, Munich, Germany

The advantages of cardiac cine MRI are often limited by its long acquisition and breath-held requirement. To overcome these limitations, we have introduced a navigator based spiral SToRM to acquire free breathing and ungated  cardiac cine MRI in a short acquisition time. Our algorithm is fully automated and does not depend on explicit binning. It gives improved image quality compared to the existing self-gated methods. Post-reconstructions, the time series can be processed to extract cardiac cycles at different respiratory phases, facilitating the estimation of anatomical and functional evaluation of the heart.

4480
Computer 108
Navigator-less Spiral SToRM for Free breathing and Ungated Cardiac CINE MRI
Abdul Haseeb Ahmed1, Ruixi Zhou2, Yang Yang2, Michael Salerno2, and Mathews Jacob1

1Electrical Engineering, University of Iowa, Iowa city, IA, United States, 2Medicine and Biomedical Engineering, University of Virginia, Charlottesville, VA, United States

This study introduces an iterative kernel low-rank algorithm to recover images in a free breathing and ungated cardiac MRI dataset. The approach relies on the manifold structure of dynamic data to recover it from highly undersampled measurements. The data is acquired using variable density spiral acquisition. An iterative kernel low-rank algorithm is introduced to estimate the manifold structure of the images, or equivalently the manifold Laplacian matrix, from central k-space regions. Unlike previous manifold regularization implementations, the iterative algorithm, coupled with the non-Cartesian acquisitions, eliminates the need for dedicated navigators to estimate the manifold Laplacian, thus improving sampling efficiency.The iterative kernel low-rank algorithm facilitates the extension of manifold regularization to navigatorless spiral acquisitions, thus improving sampling efficiency. This algorithm provides improved reconstruction compared to the state of the art methods.

4481
Computer 109
Motion Correction with a Model Target (MoCoMo): A universal approach for quantitative MRI?
Fotios Tagkalakis1, Kanishka Sharma1, Susmita Basak1, Christopher Kelly2, David Shelley3, Irvin Teh2, Jehill Parikh4, Peter Thelwall4, Neil Sheerin5, and Steven Sourbron1

1Leeds Imaging Biomarkers Group, Biomedical Imaging Science Department, University of Leeds, Leeds, United Kingdom, 2Leeds Institute of Cardiovascular and Metabolic Medicine, University of Leeds, Leeds, United Kingdom, 3Advanced Imaging Centre, University of Leeds, Leeds, United Kingdom, 4Newcastle Magnetic Resonance Centre, Newcastle University, Newcastle, United Kingdom, 5Institute of Cellular Medicine, Newcastle University, Newcastle, United Kingdom

Motion correction with a model-target (MoCoMo) has been used in DCE-MRI to overcome the problem of changes in image contrast, but the method applies in principle to any other quantitative MRI method. The aim of this study is to demonstrate this hypothesis by applying the algorithm to renal DCE, DTI, T1 and T2-mapping in human subjects. The results show that MoCoMo is effective in removing even major motion effects in all 4 modalities and does not affect data where no motion is present. We conclude that MoCoMo is a suitable candidate for universal motion correction across all functional MRI modalities. 

4482
Computer 110
Free Breathing Radial Magnetic Resonance Elastography
Joseph L Holtrop1, Stephan Kannengiesser2, Ralf B Loeffler1, Ruitian Song1, and Claudia M Hillenbrand1

1Diagnostic Imaging, St. Jude Children's Research Hospital, Memphis, TN, United States, 2Siemens Healthcare, Erlangen, Germany

Liver magnetic resonance elastography (MRE) to this point has clinically relied on using breath holds to produce reliable artifact free images. Here we present initial work adapting recent advances in motion compensated abdominal imaging for use in MRE. Specifically, we take advantage of a golden angle radial sampling scheme combined with a self-navigation approach for motion correction to perform free breathing MRE of the liver. Resulting images show enhanced detail compared to the standard breath hold technique while producing comparable image stiffness values.

4483
Computer 111
Deformable slice-to-volume registration for respiratory motion correction in abdominal and in-utero MRI
Alena Uus1, Tong Zhang1, Laurence Jackson1, Mary Rutherford1, Joseph V. Hajnal1, and Maria Deprez1

1King's College London, London, United Kingdom

This work introduces deformable slice-to-volume registration (DSVR) integrated into super-resolution reconstruction framework for correction of respiratory motion MRI. Using the initial estimation of respiratory motion as an input this method allows reconstruction of high-resolution volumes for specific respiratory positions using all slices. Based on diffeomorphic free-form deformation model, DSVR provides robust registration of deformable objects as well as out-of-plane motion correction. The feasibility of the method was successfully evaluated on a ‘motion-corrupted’ phantom and a free-breathing in-utero MRI scan. The results also indicated that the accuracy of spatial features in reconstructed volume is directly defined by the initial motion estimation.

4484
Computer 112
Groupwise Non Rigid Registration For Temporal Myocardial Arterial Spin Labeling Images
Veronica Aramendía-Vidaurreta1, Pedro Macías-Gordaliza2,3, Marta Vidorreta4, Rebeca Echeverria-Chasco1, Gorka Bastarrika1, Arrate Muñoz-Barrutia2,3, and María Fernández-Seara1

1Radiology, Clínica Universidad de Navarra, Pamplona, Spain, 2Universidad Carlos III de Madrid, Madrid, Spain, 3Instituto de Investigación Sanitaria Gregorio Marañón, Madrid, Spain, 4Siemens Healthineers, Madrid, Spain

Arterial Spin Labeling (ASL) enables quantitative measurement of myocardial blood flow (MBF) by averaging over multiple ASL pairs providing a voxelwise map in units of milliliters of blood per gram of tissue per minute (ml/g/min). However, its estimation accuracy in free breathing acquisitions depends critically on the quality of the image registration algorithm. In this work, a groupwise non-rigid registration method with a similarity measure based on Principal Component Analysis (PCA) was applied to ASL images of the heart acquired during free breathing. The method was compared against a pair-wise registration algorithm provided by the advanced normalization tools software (ANTs). Results demonstrate the feasibility of using PCA-groupwise for temporal ASL image registration.

4485
Computer 113
Reconstruction-based Super-Resolution for High-Resolution Abdominal MRI: A Preliminary Study
Michael Ebner1,2, Premal A Patel1, David Atkinson3, Lucy Caselton3, Stuart Taylor3, Alan Bainbridge4, Sebastien Ourselin1,2, Manil Chouhan3, and Tom Vercauteren1,2

1Wellcome / EPSRC Centre for Interventional and Surgical Sciences, University College London, London, United Kingdom, 2School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, 3Centre for Medical Imaging, University College London, London, United Kingdom, 4Department of Medical Physics, University College London Hospitals NHS Trust, London, United Kingdom

Magnetic resonance (MR) cholangio-pancreatography (MRCP) is an established specialist method for imaging the upper abdomen and biliary/pancreatic ducts. Due to limitations of either MR image contrast or low through-plane resolution, patients may require further evaluation with contrast-enhanced computed tomography (CT) images. However, CT fails to offer the high tissue-ductal-vessel contrast-noise ratio available on T2-weighted MRI. MR Super-Resolution Reconstruction (SRR) frameworks can provide high-resolution visualizations from multiple low through-plane resolution single-shot T2-weighted (SST2W) images as currently used during MRCP studies. Here, we investigate the clinical potential of using additional SST2W acquisitions in multiple directions with SRR for higher diagnostic yield.

4486
Computer 114
Self-navigated bulk motion detection  for feed and wrap renal dynamic radial VIBE DCE-MRI
Jaume Coll-Font1,2, Onur Afacan1,2, Alto Stemmer3, Richard S. Lee2,4, Jeanne S. Chow1,2, Simon K. Warfield1,2, and Sila Kurugol1,2

1Radiology, Boston Children's Hospital, Boston, MA, United States, 2Harvard Medical School, Boston, MA, United States, 3Siemens Healthcare GmbH, Erlangen, Germany, 4Urology, Boston Children's Hospital, Boston, MA, United States

Dynamic Radial VIBE (DRV) DCE-MRI allows to image with sufficient spatio-temporal resolution for functional imaging of kidneys. However, fast movements of babies during the scan corrupt individual lines in k-space and severely compromise the quality of the reconstructed images and limits the clinical utility of non-sedated imaging. In this work, we evaluate a self-navigated bulk motion detection approach to identify these corrupted lines. We applied this approach on non-sedated infants undergoing feed-and-wrap DCE-MRI with DRV. Our results show that this approach correctly identifies the bulk motion and allows for post-processing correction of the DCE absorption curves.

4487
Computer 115
Motion Correction Resolved for MRI via Multi-Tasking: A Simultaneous Reconstruction and Registration Approach
Veronica Corona1, Noémie Debroux1, Angelica I. Aviles-Rivero2, Guy Williams3, Martin J. Graves4, Carole Le Guyader5, and Carola-Bibiane Schoenlieb1

1Department of Applied Mathematics and Theoretical Physics, University of Cambridge, Cambridge, United Kingdom, 2Department of Pure Mathematics and Mathematical Statistics, University of Cambridge, Cambridge, United Kingdom, 3Wolfson Brain Imaging Centre, Department of Clinical Neurosciences, University of Cambridge, Cambridge, United Kingdom, 4Cambridge University Hospitals, Department of Radiology, University of Cambridge, Cambridge, United Kingdom, 5INSA Rouen, Laboratoire de Mathématiques, Normandie Université, Saint-Étienne-du-Rouvray, France

The prolonged time required to form an MR image continues to impose different challenges at both theoretical and clinical levels.  With this motivation in mind, this work addresses a central topic in MRI, which is how to correct the motion problem, through a new multitask optimisation framework. The significance is that by tackling the reconstruction and registration tasks $$$-$$$ simultaneously and jointly $$$-$$$ one can exploit their strong correlation reducing error propagations and resulting in a significant motion correction. The clinical potentials of our approach are reflected in having higher image quality with fewer artefacts whilst keeping fine details. We evaluate our approach through a set of quantitative and qualitative experimental results.

4488
Computer 116
Free-breathing MRI of the upper abdomen assisted by motion modelling
Robert I Johnstone1,2, David Atkinson3, Manil Chouhan3, Ricky A Sharma4, and Jamie R McClelland1

1Centre for Medical Image Computing, University College London, London, United Kingdom, 2Medical Physics, Guy's and St. Thomas' NHS Foundation Trust, London, United Kingdom, 3Centre for Medical Imaging, University College London, London, United Kingdom, 4NIHR University College London Hospitals Biomedical Research Centre, University College London, London, United Kingdom

This study demonstrates the use of motion modelling and super resolution reconstruction (SRR) to produce an isotropic 3D image of the upper abdomen during free breathing.

 

Sagittal and coronal 6 mm 2D slices are acquired throughout the volume of interest. The slices are repeated with sub-voxel offsets to facilitate SRR. An interleaved navigator slice is also acquired.

 

The navigator slice is processed with non-rigid registration and principal component analysis, to give two motion surrogate signals. These signals are used to control the motion model. The motion model and the SRR are jointly optimised using an iterative scheme.


4489
Computer 117
A segmented ultra-short echo (UTE) sequence equipped with robustness to respiratory motion
Cihat Eldeniz1, Uday Krishnamurthy2, and Hongyu An1

1Mallinckrodt Institute of Radiology, Washington University in St. Louis, St. Louis, MO, United States, 2Siemens Healthineers, St. Louis, MO, United States

Synopsis

4490
Computer 118
Effects of Image Registration in Dynamic Contrast-Enhanced MRI of the TMJ
Lea Sjurine Starck1,2, Erling Andersen2,3, Ondřej Macíček4, Oskar Angenete5,6, Thomas Augdal7, Erlend Hodneland2,8, Radovan Jiřík4, Karen Rosendahl9,10, and Renate Grüner1,2,7

1Dept. of Physics and Technology, University of Bergen, Bergen, Norway, 2Mohn Medical Imaging and Visualization Centre, Haukeland University Hospital, Bergen, Norway, 3Dept. of Clinical Engineering, Haukeland University Hospital, Bergen, Norway, 4Institute of Scientific Instruments, Czech Academy of Sciences, Brno, Czech Republic, 5Dept. of Radiology and Nuclear Medicine, St. Olav hospital HF, Trondheim, Norway, 6Dept. of Circulation and Medical Imaging, Norwegian university of Science and Technology, Trondheim, Norway, 7Dept. of Radiology, University Hospital of North Norway, Tromsø, Norway, 8Norse, Bergen, Norway, 9Dept. of Radiology, Haukeland University Hospital, Bergen, Norway, 10Dept. of Clinical Sciences, University of Bergen, Bergen, Norway

The effect of elastic and affine motion correction in dynamic contrast enhanced MRI ofthe temporomandibular joints in children is investigated. Imaging in children is particularly difficultdue to motion. This hampers DCE-MRI and pharmacokinetic estimations for their potentialdiagnostic value in these children with Juvenile Idiopathic Arthritis with possible TMJ involvement.The relative enhancement curves obtained with different motion correction approaches arecompared with the curves calculated with the Gamma Capillary Transit Time model. It is found thatwhen image registration is applied, a greater number of participants can be analysed. The elasticmotion correction approach outperforms the affine approach.

4491
Computer 119
Magnetic tracking of ECG sensors for respiratory motion correction
Benjamin Roussel1,2, Joris Pascal3, Jacques Felblinger1,2, and Julien Oster1,2

1Université de Lorraine, Nancy, France, 2U1254, INSERM, Nancy, France, 3FHNW/HLS/IMA, FHNW/HLS/IM2, University of Applied Sciences and Arts Northwestern Switzerland FHNW, Muttenz, Switzerland

Monitoring the respiration motion is a crucial step for motion correction. We propose a magnetic tracking system, using a magnetic sensor and a Helmholtz coil as the magnetic field source. By comparing the sensed magnetic fields with theoretical values under the dipole approximation, we were able to locate sensors placed on a subject’s chest and track their motion during breathing. With a sub-centimeter resolution and the current sources of imprecision being identified, we are confident this method can be a viable solution for accurate motion monitoring in MRI, especially by using the magnetic fields generated by the gradient coils.

4492
Computer 120
Iterative Static Motion Compensated(IS-MoCo) Reconstruction: application to high resolution lung imaging
Xucheng Zhu1,2, Kevin M. Johnson3, Michael Lustig4, and Peder E.Z. Larson1

1Radiology, University of California San Francisco, San Francisco, CA, United States, 2Bioengineering, University of California San Francisco, San Francisco, CA, United States, 3Department of Medical Physics, University of Wisconsin, Madison, Madison, WI, United States, 4Department of Electrical Engineering and Computer Sciences, UC Berkeley, Berkeley, CA, United States

High resolution 3D MRI thoracic and abdominal MRI is always challenging, due to long acquisition time and susceptibility to subject motion. We proposed a novel reconstruction method, named Iterative Static Motion Compensated(IS-MoCo) reconstruction, to compensate motion affects during the reconstruction instead of gating. The proposed method is applied to high resolution free breathing lung imaging, outperforms widely used motion correction strategies with higher SNR and less residual motion artifacts.

4493
Computer 121
Respiratory Motion Corrected GROG based L+S Reconstruction for Free Breathing Golden-Angle Radial MRI
Ahmad Hussain1, Faisal Najeeb1, Ibtisam Aslam1, Hammad Omer1, and Mujahid Nisar1

1Electrical Engineering, Comsats University Islamabad, Islamabad, Pakistan

Respiratory motion during MRI scan causes inconsistencies in the acquired k-space data providing strong blurring artifacts in the reconstructed images. In this work, a new method ( respiratory motion corrected GROG followed by L+S reconstruction for free breathing Golden-Angle Radial DCE-MRI) is presented.The proposed method is tested on 3-T free-breathing Golden angle radial DCE liver MRI data. The proposed method is compared with the conventional L+S reconstruction model. The proposed method provides 90% improvement in Artefact Power and 42% in RMSE as compared to conventional L+S reconstruction at acceleration factor 8.

4494
Computer 122
Impact of registration on multi-parametric breast MRI data and parameters: Qualitative and Quantitative Assessment
Snekha Thakran1, Subhajit Chatterjee1,2,3, Dinil Sasi1, Ayan Debnath1,4, Rupsa Bhattacharjee1, Rakesh Kumar Gupta5, and Anup Singh1,6

1Centre for Biomedical Engineering, Indian Institute of Technology Delhi, New Delhi, India, 2Department of Computer Science and Engineering, Indian Institute of Technology Delhi, New Delhi, India, 3C-DOT India, Delhi, India, 4Center for Magnetic Resonance & Optical Imaging, University of Pennsylvania, Philadelphia, PA, PA, United States, 5Department of Radiology, Fortis Memorial Research Institute, Gurgaon, Haryana, New Delhi, India, 6Department of Biomedical Engineering, All India Institute of Medical Science, New Delhi, India

Multi-parametric(mp)-MRI data such as conventional MRI, DCE-MRI, DWI, etc. are routinely acquired for breast cancer patients. Any motion during mp-MRI data acquisition can affect qualitative as well as quantitative mp-MRI results. In this study, impact of registration on mp-MRI data as well as on quantitative parameters was evaluated qualitatively and quantitatively. Study included mp-MRI data of 40 patients with breast cancer. B-spline based registration performed better than Affine and SyN. It showed highest dice-coefficient, correlation coefficient. It also provided better histograms of quantitative maps and provided lowest sum-of-squared error in signal-intensity curves from ROI at edge and center of lesion.

4495
Computer 123
MR-assisted PET motion correction improves tumor-to-background and contrast-to-noise ratios in a phantom study with ground truth reference
Sihao Chen1, Cihat Eldeniz1, Richard Laforest1, and Hongyu An1

1Washington University in St. Louis, Saint Louis, MO, United States

Respiratory motion leads to signal blurring and reduced tumor-to-background (TBR) and contrast to noise (CNR) ratios. As a result, it can severely affect the detectability of lesions in PET imaging.1,2 Simultaneous PET/MR imaging uniquely allows for MR assisted motion correction in PET imaging.3 In this study, we have demonstrated that the MR assisted PET motion correction significantly improves both tumor-to-background and contrast-to-noise ratios, leading to better lesion detection.

4496
Computer 124
Blind Sparsity Based Motion Estimation and Correction Model for Arbitrary MRI Sampling Trajectories
Anita Möller1, Marco Maass1, Tim Jeldrik Parbs1, and Alfred Mertins1

1Institute for Signal Processing, Universität zu Lübeck, Lübeck, Germany

A blind retrospective MRI motion estimation and compensation algorithm is designed for arbitrary sampling trajectories. Using the idea of natural images being sparsely representable, the algorithm is based on motion estimation between a motion corrupted image and it’s sparse representative. Therefore, rigid motion models are designed and used in gradient descent methods for image quality optimization. As the motion estimation and compensation work on arbitrary real valued sampling coordinates, the algorithm is capable for all trajectories. Image reconstruction is performed by computationally efficient gridding. The exact motion estimation results are shown for PROPELLER and radial trajectory simulation.

4497
Computer 125
Self-Gated Pulmonary Embolism Imaging with Multi-Slice Golden-Angle Radial bSSFP
Alexander Fyrdahl1,2, Roberto Vargas-Paris3,4, Koshiar Medson3,5, Sven Nyrén1,5, Martin Ugander1,2, Peter Lindholm3,5, and Andreas Sigfridsson1,2

1Department of Molecular Medicine and Surgery, Karolinska Institutet, Stockholm, Sweden, 2Department of Clinical Physiology, Karolinska University Hospital, Stockholm, Sweden, 3Department of Physiology and Pharmacology, Karolinska Institutet, Stockholm, Sweden, 4Department of Radiology, Karolinska University Hospital, Stockholm, Sweden, 5Department of Thoracic Radiology, Karolinska University Hospital, Stockholm, Sweden

Free-breathing, non-contrast bSSFP MRI has shown great potential for imaging of pulmonary embolisms in patients with contraindication for contrast-enhanced computed tomography angiography. While the free-breathing approach is convenient, it limits the possibility for multiplanar reformatting which otherwise could aid in visualizing the pulmonary vasculature. In this work, we propose a methodology for deriving a motion signal from the free-breathing data and we incorporate this signal in the reconstruction pipeline to obtain a slice-aligned image stack from which multiplanar reformatting can be performed.


System Imperfections

Exhibition Hall
Thursday 8:15 - 9:15
 Acquisition, Reconstruction & Analysis

4498
Computer 126
Minimal Linear Networks for MR Image Reconstruction
Gilad Liberman1 and Benedikt A Poser1

1Faculty of Psychology and Neuroscience, Maastricht University, Maastricht, Netherlands

We propose minimal linear networks (MLN) for MR image reconstruction that employ complex-valued, axis-dependent and fully- and neighborhood-connected layers with shared and independent weights, Their topology is restricted to the minimum required by the MR-physics, without nonlinear activation layers. The suggested MLN perform well in reconstructing imaging data acquired under challenging real-world imaging conditions, specifically an Arterial Spin Labeling perfusion experiment with spiral sampling at 7 Tesla. Despite the strong B0 field inhomogeneities at 7T, artifact-free images are obtained that are capable of resolving the minute perfusion signal changes. The results show that even without nonlinear activation and higher-order image manifold description as used by others, deep-learning algorithms and framework, and learning from large realistic datasets, can play a significant role in the success of image reconstruction.

4499
Computer 127
Correcting breast MRI with a generic B-1(+) template for T-1 map calculation
Lieke van den Wildenberg1, Erwin Krikken1, Jannie P. Wijnen1, Josien P. W. Pluim2, Dennis W. J. Klomp1, and Michael J. van Rijssel1

1Center for Image Sciences, UMC Utrecht, Utrecht, Netherlands, 2Biomedical Engineering, Eindhoven University of Technology, Eindhoven, Netherlands

In 7T breast MRI, the use of local transmit coils causes an inhomogeneous B1+ field, decaying towards the pectoral muscle. This leads to differences in image contrast throughout the breasts and in dynamic contrast enhanced (DCE) MR images it has a direct influence on the enhancement kinetic curves. Therefore a correction is necessary. We used B1+ simulations to generate a template to correct the images, because the dynamic range of measured B1+ maps is often insufficient. We validated the template on eleven volunteers. T1-maps were calculated using the generic template as a first step of correcting the DCE images.

4500
Computer 128
Removing bias and increasing dynamic range in DREAM flip angle mapping at 7T
Gunther Helms1,2, Hampus Olsson1, and Mads Andersen2,3

1Clinical Sciences Lund, Medical Radiation Physics, Lund University, Lund, Sweden, 2National 7T Facility, Lund University, Lund, Sweden, 3Philips Danmark A/S, Philips Healthcare, Copenhagen, Denmark

DREAM is an ultra-fast multi-slice B1+-mapping technique based on the single-shot STEAM sequence. To study noise and bias related to slice-profiles, DREAM B1+-maps at 3.75mm resolution were acquired at 7T in phantoms and in human brain with nominal flip angles (FA) between 20° and 90° of the two STEAM preparation pulses. B1+ was decreasing at actual FAs above 50°; noise became apparent at actual FAs below 20° reducing dynamic range. By varying the preparation FA, this reliable range (20°<FA<50°) is shifted over a B1+ range from 20% to 250%. The FA map is constructed from overlapping B1+ maps after thresholding.

4501
Computer 129
Optimal Flip Angle Range for B1+ Mapping at 3T with Slice Profile Correction Using a Dual Angle EPI Sequence
Gabriela Belsley1, Damian J. Tyler1, Matthew D. Robson1,2, and Elizabeth M. Tunnicliffe 1

1Oxford Centre for Clinical Magnetic Resonance Research, Division of Cardiovascular Medicine, Radcliffe Department of Medicine, University of Oxford, Oxford, United Kingdom, 2Perspectum Diagnostics, Oxford, United Kingdom

Mapping B1+ inhomogeneity, using commonly available pulse sequences, is essential for widespread, accurate determination of T1 using variable flip angle methods. We investigated the accuracy of B1+ mapping with different flip angles (FA) using the double angle method with a 2D multi-slice GRE-EPI sequence. At lower FAs, we found that B1+ accuracy is affected by SNR, whereas the extent of B1+ inhomogeneities imposes an upper limit on the FAs that can be employed. For a B1+ inhomogeneity of ±40% and a SNR of 29 at 30°, the optimal FA pairs were found to lie between 43°/86° and 74°/148°. 

4502
Computer 130
Retrospective transmission field (B1+) and sensitivity profile (B1-) correction for transceive surface RF coils: an empirical solution for RARE
Paula Ramos Delgado1, Andre Kuehne2, Ludger Starke1, Jason M. Millward1, Joao Periquito1, Thoralf Niendorf1,3, Sonia Waiczies1, and Andreas Pohlmann1

1Berlin Ultrahigh Field Facility (B.U.F.F), Max Delbrück Center for Molecular Medicine, Berlin, Germany, 2MRI.tools GmbH, Berlin, Germany, 3Experimental and Clinical Research Center, a joint cooperation between the Charité Medical Faculty and the Max Delbrück Center for Molecular Medicine in the Helmholtz Association, Berlin, Germany

Improving the low signal-to-noise ratio (SNR) inherent to emerging MRI methods such as fluorine MRI is challenging. To enhance sensitivity, SNR-efficient pulse sequences such as RARE and cryogenically-cooled surface RF coils (CRP) are used. Transceive surface RF coils show variation in the excitation field (B1+), impairing quantification. To compensate, previous studies have used an analytical signal intensity equation to perform a retrospective B1+-correction. However, this is unfeasible for RARE due to the absence of such an equation. To overcome this challenge, we propose and validate a numerical method using experimental data acquired with a volume resonator (reference) and a 1H-CRP.

4503
Computer 131
Dynamic Decoupling for Simultaneous Transmission and Acquisition in MRI
Bilal Tasdelen1,2, Alireza Sadeghi-Tarakameh1,2, Ugur Yilmaz2, and Ergin Atalar1,2

1Department of Electrical and Electronics Engineering, Bilkent University, Ankara, Turkey, 2National Magnetic Resonance Research Center (UMRAM), Ankara, Turkey

In order to use simultaneous transmission and acquisition in clinical MRI for living subjects, robustness to load and environmental changes has to be established, especially for uncooperative subjects. High isolation can be achieved with active cancellation methods, but maintaining it over a long time is a challenge. A look-up table based method is proposed with a smart search algorithm that enables fast dynamic decoupling of transmit/receive coils using an active decoupling circuit. Experiments with a birdcage coil used as a transceiver show that maintaining  >80 dB isolation is possible even under the presence of load variation.


4504
Computer 132
Simple and effective trajectory estimation for image reconstruction of accelerated k-space acquisition on non-rectangular periodic trajectories
Kazi Rafiqul Islam1 and Jingxin Zhang1

1School of Software and Electrical Engineering, Swinburne University of Technology, Melbourne, Australia

Precise coordinates of trajectories are essential for image reconstruction of k-space data acquired from non-rectangular trajectories, and measurement of the trajectories often requires prescan calibration that complicates the process. This abstract presents a simple and effective method to estimate the coordinates of non-rectangular periodic trajectories from normal scan data and demonstrates its efficacy in image reconstruction of in vivo scan data acquired from ZIGZAG trajectory.

4505
Computer 133
Absolute Phase of Radio-frequency Transmit Field B1+ for a Dual Transmit Coil System
Jinghua Wang1, Yu Ding2, Christopher Sica3, and Qing X Yang3

1Radiology, University of Cincinnati, Cincinnati, OH, United States, 2United Imaging Healthcare America, Inc, Houston, TX, United States, 3Radiology, Penn State University, Hershey, PA, United States

The spatial absolute phase information is important in various stages of MRI scanning, such as parallel image reconstruction, the combination of MR image or MR spectroscopy from each element of multiple receivers and exploration of new contrast and biomarker. Currently, the absolute phase of a transmit field can only be roughly estimated as half of the transceiver phase. This method is not only inaccurate but also not applicable for a transceiver coil.  Thus, the accurate estimation of the absolute phase for an arbitrary RF coil system is an unsolved problem and an unmet need of the MR society. Here we propose a new approach to solve this old problem.

4506
Computer 134
Evaluation of the Uniform Combined Reconstruction (UNICORN) Algorithm for Improving 7T Knee MRI Uniformity
Venkata Veerendranadh Chebrolu1, Peter Kollasch1, Vibhas Deshpande2, John Grinstead3, Benjamin Howe4, Matthew Frick4, Andrew J. Fagan4, Thomas Benner 5, Robin M. Heidemann 5, Joel Felmlee 4, and Kimberly K Amrami4

1Siemens Healthineers, Rochester, MN, United States, 2Siemens Healthineers, Austin, TX, United States, 3Siemens Healthineers, Portland, OR, United States, 4Department of Radiology, Mayo Clinic, Rochester, MN, United States, 5Siemens Healthineers, Erlangen, Germany

MR image intensity non-uniformity is often observed at 7T. A novel algorithm termed ‘Uniform Combined Reconstruction’ (UNICORN) was developed recently to correct for intensity non-uniformity in MR images without the use of a calibration/reference scan. In this work, 3 fellowship trained musculoskeletal radiologists with cumulative experience of 42 years evaluated the efficacy of UNICORN in 33 7T musculoskeletal MRI volumes. The uniformity, contrast, signal-to-noise-ratio and overall image quality were evaluated. Without the use of a reference scan, UNICORN was rated to provide better image uniformity, contrast and overall image quality than the N4 bias-field correction algorithm at 7T.

4507
Computer 135
Investigation of the Cost Function for Joint Estimation of Object and B0
Franz Patzig1, Bertram Wilm1, and Klaas Pruessmann1

1University of Zurich and ETH Zurich, Zurich, Switzerland

Due to their short read-out time single-shot techniques are frequently used for several imaging modalities but they are prone to static B0 off-resonance artifacts. To avoid separately acquired field maps joint estimation of the object and the B0 map has been proposed as a potential solution alternating between updating an object and a field map guess. A measure to compare cost functions is introduced and two different joint estimation cost functions are investigated whereby a new cost function in image space is suggested. It shows its potential if only a less reliable B0 map guess is given.

4508
Computer 136
Coil-induced phase removal during gradient delay estimation
Chao Wang1, Gaojie Zhu1, Hai Luo1, Bei Lv1, Xiang Zhou1, and Ziyue Wu1

1Alltech medical system, Chengdu, China

Gradient delay can lead to severe artifacts in radial imaging. While several methods have been proposed to correct the linear phase caused by gradient delay, no publications have mentioned the impact of coil sensitivity phase during the estimation of gradient delay to our knowledge. This work reports the impact of this factor and presents a simple method to remove the coil-induced phase during the gradient delay estimation. Both phantom and in-vivo test results are provided to demonstrate the effectiveness of this method.  

4509
Computer 137
Magnetic field estimation with ultrashort echo time (UTE) imaging.
Jiazheng Zhou1,2, Ali Aghaeifer1,2,3, Jonas Bause1,2,3, Alexander Loktyushin1,4, Gisela Hagberg1,3, and Klaus Scheffler1,3

1High-Field Magnetic Resonance, Max Planck Institute for Biological Cybernetics, Tübingen, Germany, 2Graduate Training Center of Neuroscience, IMPRS, University of Tübingen, Tübingen, Germany, 3Biomedical Magnetic Resonance, University Hospital Tübingen (UKT), Tübingen, Germany, 4Empirical Inference, Max Planck Institute for Intelligent Systems, Tübingen, Germany

We have used UTE sequence to obtain the subject-specific susceptibility distribution, which was then used to simulate motion-induced B0 change at two head positions. A Fourier-based dipole-approximation method was used to map susceptibility to B0. We have evaluated the simulation results against the measured B0 at the same positions and observed a good agreement between the simulated and real data.

4510
Computer 138
Correction for Geometric Distortion in Bipolar Gradient Echo Images from $$$B_0$$$ Field Variations
Korbinian Eckstein1, Siegfried Trattnig1, and Simon Daniel Robinson1

1High Field Magnetic Resonance Center, Department of Biomedical Imaging and Image-Guided Therapy, Medical University of Vienna, Vienna, Austria

In bipolar multi-echo gradient echo imaging, signal is acquired during positive and negative readout gradients, giving an efficiency advantage over monopolar imaging in which no signal is acquired during “fly-back”/rewind periods. This increased acquisition efficiency allows higher resolution, shorter echo spacing or increased SNR. In bipolar acquisitions, however, $$$B_0$$$-related distortion along the readout axis occurs in opposite directions for odd and even echoes, leading to blurring when images from echoes are combined. We show that a simple unwarping scheme, based on $$$B_0$$$ field maps derived from the multi-echo data themselves, is effective in correcting this effect in multi-echo SWI.

4511
Computer 139
Routine B0 eddy current measurements with TOPPE for more robust spiral imaging
Jon-Fredrik Nielsen1, Gopal Nataraj2, Jeffrey Fessler1, and Douglas Noll1

1University of Michigan, Ann Arbor, MI, United States, 2Memorial Sloan Kettering Cancer Center, New York, NY, United States

Spiral imaging is an SNR- and time-efficient alternative to conventional cartesian MRI, but is relatively sensitive to gradient system imperfections. Unfortunately, measuring the k-space trajectory and B0 eddy currents for a particular spiral readout is cumbersome and not routinely performed. We propose to leverage the TOPPE development environment for rapid pulse sequence prototyping to easily measure both k-space trajectory and B0 eddy currents using “pencil-beam” acquisitions. To demonstrate this setup, we obtained k-space and B0 measurements of a pair of spiral-in and spiral-out readouts. We show that compensating for B0 eddy currents can improve image quality, and that TOPPE provides a convenient platform for these types of measurements.

4512
Computer 140
Cross-vendor spiral gradient calibration using TOPPE and Pulseq
Jon-Fredrik Nielsen1, Douglas Noll1, and Sairam Geethanath2

1University of Michigan, Ann Arbor, MI, United States, 2Columbia University, New York, NY, United States

Spiral imaging is fast and SNR-efficient, but is relatively sensitive to gradient system imperfections. Unfortunately, these imperfections are generally not known, and can furthermore be expected to vary across different scanner vendor platforms. This complicates multi-site, multi-vendor studies that can benefit from rapid spiral imaging, e.g., those involving fMRI. Here we demonstrate that it is in fact possible to characterize and directly compare spiral gradient performance across two major vendors (GE and Siemens), using the TOPPE and Pulseq frameworks for rapid pulse sequence prototyping. Our observations indicate that B0 eddy currents are substantial on both vendor platforms, and underscore the need for measuring and correcting for B0 effects in spiral imaging.

4513
Computer 141
Multiparametric evaluation of geometric distortions in stereotactic MR imaging at 1.5 and 3 Tesla with a plexiglass phantom: towards practical recommendations for clinical imaging protocols
Gizem Temiz1,2, Fernando Pérez-García1,2, Catherine Jenny3, Stéphane Lehéricy1,4, Marguerite Cuttat 3, Didier Dormont4, Damien Galanaud1,4, Chales Valery5, Carine Karachi1,5, Romain Valabregue 2, Sara Fernandez-Vidal1,2, Nadya Pyatigorskaya4, and Eric Bardinet1,2

1Institut du Cerveau et de la Moelle épinière, ICM, Inserm U 1127, CNRS UMR 7225, Sorbonne Université, F-75013, Paris, France, 2Centre de Neuro-Imagerie de Recherche, CENIR, Paris, France, 3Department of Radiotherapy, Medical Physics Unit, AP-HP Pitié-Salpêtrière Hospital, Paris, France, 4Department of Neuroradiology, AP-HP Pitié-Salpêtrière Hospital, Paris, France, 5Department of Neurosurgery, AP-HP Pitié-Salpêtrière Hospital, Paris, France

Accurate MRI-based targeting is a critical issue for stereotactic surgery. Therefore, geometric distortions need to be evaluated for any pre-operative MR imaging protocol. In this study, we investigated MRI protocols used in Deep Brain Stimulation and Gamma Knife radiosurgery, and focused on the influence of 5 factors on the geometric distortions, at 1.5T and 3T, for 3D T1-weighted and 3D FLAIR images. We found that in order to minimize geometric distortions in stereotactic imaging operator training, careful centering in the MR scanner and systematic activation of constructor’s distortion correction filter are essentials.

4514
Computer 142
MRI quality data assessment in the Italian IRCCS advanced neuroimaging network using ACR phantoms
Fulvia Palesi1, Anna Nigri2, Domenico Aquino2, Ruben Gianeri2, Alice Pirastru3, Marcella Laganà3, Laura Biagi4, Michela Tosetti4, Maria Grazia Bruzzone2, Claudia A.M. Gandini Wheeler-Kingshott5,6,7, and The Italian IRCCS advanced neuroimaging network8

1Neuroradiology Unit, Brain MRI 3T Research Center, IRCCS Mondino Foundation, Pavia, Italy, 2Neuroradiology, Fondazione I.R.C.C.S. Istituto Neurologico Carlo Besta, Milan, Italy, 3IRCCS Fondazione Don Carlo Gnocchi, Milano, Italy, 4IRCCS Fondazione Stella Maris, Pisa, Italy, 5Queen Square MS Centre, Department of Neuroinflammation, UCL Queen Square Institute of Neurology, Faculty of Brain Sciences, University College London, London, United Kingdom, 6Department of Brain and Behavioural Sciences, University of Pavia, Pavia, Italy, 7Brain MRI 3T Research Center, IRCCS Mondino Foundation, Pavia, Italy, 8The Italian IRCCS advanced neuroimaging network, Milan, Italy

Generating big-data is becoming imperative with the advent of machine learning. Neuroimaging networks respond to this need. Italian Research Neurological Institutes have formed an advanced neuroimaging network to develop protocols for multisite studies. The present work reports on ACR phantom data across sites and evaluates accuracy and longitudinal reproducibility of: uniformity and ghosting, geometric accuracy, slice thickness, high-contrast and low-contrast object detectability. Our findings show that uniformity, geometric accuracy, low-contrast object detectability are measures that failed at some sites. We intervened to correct these issues improving protocol quality and scanner stability, establishing levels of precision relevant for future multicentre studies in quantitative imaging.

4515
Computer 143
Retrospective 3D spiral trajectory correction: Exponential decay model vs. GIRF
Tobias Speidel1, Anja Schwarz2, Kilian Stumpf3, and Volker Rasche4

1Core-Facility Small Animal Imaging (CF-SANI), Ulm University, Ulm, Germany, 2Ulm University, Ulm, Germany, 3Ulm University Medical Center, Ulm, Germany, 4Experimental Cardiovascular MRI (ExCaVI), Ulm University Medical Center, Ulm, Germany

Uncorrected gradient imperfections lead to degrading image artifacts, especially in the case of demanding 3D trajectories. Different post-processing methods have been introduced to compensate for the real-time behaviour of the gradient system. This work shows that gradient waveform deviations can be vastly and nearly equally corrected using an exponential decay model as well as the gradient impulse response function. Both approaches were applied to a pure 3D spiral-like trajectory (Seiffert's spiral), achieving a comparable enhancement in image quality.

4516
Computer 144
Retrospective gradient delay correction in multi-shot multi-echo rosette acquisition
Alexey V. Dimov1, Nanyque A. Boyd2, Keigo Kawaji2,3, and Timothy J. Carroll1

1Radiology, University of Chicago, Chicago, IL, United States, 2Biomedical Engineering, Illinois Institute of Technology, Chicago, IL, United States, 3Medicine, University of Chicago, Chicago, IL, United States

Rosette k-space sampling is an attractive tool for a variety of applications such QSM, DCE, etc.  However, as many non-Cartesian acquisition schemes, rosette is highly susceptible to the system-specific gradient delays. We present a robust technique utilizing intrinsic symmetries of multi-shot rosettes, which allows to reconstruct images with minimal artifacts due to misalignment of the k-space points.

4517
Computer 145
Imaging Beyond the Homogeneous Radius in Clinical Magnets
Nadine Luedicke Dispenza1, Yanitza Marie Rodriguez2, Robert Todd Constable2,3, and Gigi Galiana2

1Biomedical Engineering, Yale University, New Haven, CT, United States, 2Radiology and Biomedical Imaging, Yale University, New Haven, CT, United States, 3Department of Neurosurgery, Yale University, New Haven, CT, United States

 At the edge of the bore, both B0 and gradient fields have nonlinear distortions which typically make imaging impossible beyond a certain FOV.  In this work, we show that undistorted imaging without significant loss of SNR can be achieved outside the typical 50 cm imaging volume using nonlinear spatial encoding techniques. Spatial distortions are corrected by incorporating the nonlinearities of the gradients and B0 field in the encoding equation, while intravoxel dephasing information counteracts spurious intensity modulations and blurring in the reconstructed images. 

4518
Computer 146
Accuracy, repeatability, and reproducibility of longitudinal relaxation rate in twelve small-animal MR systems.
John C Waterton1,2, Catherine DG Hines3, Paul D Hockings4,5, Iina Laitinen6, Sabina Ziemian7, Simon Campbell8, Michael Gottschalk9, Claudia Green7, Michael Haase8, Katja Hoffmann6, Hans-Paul Juretschke6, Sascha Koehler10, William Lloyd1, Yanping Luo11, Irma Mahmutovic Persson9, James PB O'Connor1, Lars E Olsson9, Geoffrey JM Parker1,2, Kashmira Pindoria8, Juergen E Schneider12, Steven Sourbron12, Denise Steinmann6, Klaus Strobel10, Sirisha Tadimalla12, Irvin Teh12, Andor Veltien13, Xiaomeng Zhang11, and Gunnar Schütz7

1University of Manchester, Manchester, United Kingdom, 2Bioxydyn, Manchester, United Kingdom, 3Merck, West Point, PA, United States, 4Antaros, Mölndal, Sweden, 5Chalmers University of Technology, Gothenburg, Sweden, 6Sanofi-Aventis, Frankfurt-am-Main, Germany, 7Bayer, Berlin, Germany, 8GlaxoSmithKline, Stevenage, United Kingdom, 9Lund University, Lund, Sweden, 10Bruker, Ettlingen, Germany, 11Abbvie, North Chicago, IL, United States, 12University of Leeds, Leeds, United Kingdom, 13Radboud university medical center, Nijmegen, Netherlands

Many translational MR biomarkers derive from measurements of R1, but evidence for between-site reproducibility of Rin small-animal MRI is lacking.  Here R1 was measured by saturation-recovery in 2% agarose phantoms with five NiCl2 concentrations in 12 magnets at 5 field strengths in 11 centres on two different occasions within 1-13 days.  R1 was analysed in three different regions of interest, giving 360 measurements in total.  Root-mean-square repeatability and reproducibility coefficients of variation were calculated.  Day-to-day repeatability was 2.3%.  Between-centre reproducibility was 1.4%.  Ni2+ relaxivity in 2% agarose was 0.66s-1mM-1 at 3T and 0.94s-1mM-1 at 11.7T.

4519
Computer 147
On the impact of slice profile and thickness definition across vendors in 2D bSSFP on SNR and T1-mapping in cardiac MRI
Jouke Smink1, Guillaume Gilbert2, Marc Kouwenhoven1, and Johan S van den Brink3

1MR Clinical Science, Philips, Best, Netherlands, 2MR Clinical Science, Philips, Montréal, QC, Canada, 3MR Clinical Excellence, Philips, Best, Netherlands

The actual slice thickness and slice profile in 2D imaging are often not taken into account when comparing SNR from different platforms. It can also have an impact in quantitative imaging such as T1-mapping. Inspired by an earlier study, we compared two definitions of slice thickness in 2D bSSFP(the workhorse in CMR): full width at 50% (FW50) and full width at 70% of maximum (FW70). The FW70 pulse definition leads to 30% thicker slices, 9-30% more SNR and it is more vulnerable to partial volume effects. These effects needs to be taken into account when comparing scans from different platforms in multi-center trials.

4520
Computer 148
Regularization of Digitally Integrated, Inductive k-Space Trajectory Measures
Jan Ole Pedersen1,2,3, Christian G. Hanson4, Rong Xue5,6, and Lars G. Hanson7,8

1Danish Research Centre for Magnetic Resonance, Hvidovre, Denmark, 2Sino-Danish Center for Education and Research, Aarhus, Denmark, 3Philips Healthcare, København SV, Denmark, 4Denmark, 5Chinese Academy of Sciences, State Key Laboratory of Brain and Cognitive Sciences, Institute of Biophysics, Beijing, China, 6University of Chinese Academy of Sciences, Sino-Danish College, Beijing, China, 7Department of Electrical Engineering, Technical University of Denmark, Center for Magnetic Resonance, Kgs Lyngby, Denmark, 8Centre for Functional and Diagnostic Imaging and Research, Copenhagen University Hospital, Danish Research Centre for Magnetic Resonance, Hvidovre, Denmark

Determining k-space trajectories inductively is conceptually simple, but rely on integration of the induced signal. Performing this integration digitally allow for higher degree of flexibility than analog integration, which is necessary to account for, e.g., refocusing RF pulses. Digital integration, however, require high bandwidth sampling of the induced signal as digitization error accumulate, making the overall approach less attractive. We show that the necessary bandwidth can be reduced by performing regularization using a gradient coil current measure.  

4521
Computer 149
Restoring Rotation Invariance of Diffusion MRI Estimators in the Presence of Missing or Corrupted Measurements
Hans Knutsson1, Evren Özarslan1, Filip Szczepankiewicz2, and Carl-Fredrik Westin1,2

1Linkoping University, Linkoping, Sweden, 2Brigham and Women's Hospital, Harvard Medical School, Boston, MA, United States

A natural requirement of estimated tissue microstructure features is that they are rotation invariant. So far, the strategy to attain rotation invariance has been to measure in as many uniformly distributed directions as can be afforded and simply compute the projections on an appropriate set of angular basis functions. However, in the presence of missing samples, this approach is sub-optimal. We show that attaching carefully chosen weights to each measurement can achieve a significantly improved rotation invariance, even in the presence of corruptions that break the isotropic sampling symmetry.


Optimization of Quantitative Mapping Techniques

Exhibition Hall
Thursday 8:15 - 9:15
 Acquisition, Reconstruction & Analysis

4522
Computer 151
The need of a varying flip angle in multi-component analysis with IR-bSSFP sequences.
Martijn Nagtegaal1, Thomas Amthor2, Peter Koken2, and Mariya Doneva2

1TU Berlin, Berlin, Germany, 2Philips Research Europe, Hamburg, Germany

A comparative analysis between IR-bSSFP and MR Fingerprinting was performed in numerical simulations for single and multi-component parameter mapping. The single component matching works for both methods, although the accuracy for T2 is better for MR Fingerprinting. The multi component matching for a constant flip angle IR-bSSFP sequence can only match to the T1* values and cannot distinguish between the underlying T1/T2 values. Using the MR Fingerprinting sequence with a varying flip angle it is possible to match to the T1/T2 components.

4523
Computer 152
Schedule design for parameter quantification in the transient state using Bayesian optimisation
Giada Fallo1,2, Matteo Cencini2,3, Pedro A. Gómez4, Davide Bacciu1, Antonio Cisternino1, Michela Tosetti2, and Guido Buonincontri2

1Dipartimento di Informatica, Università di Pisa, Pisa, Italy, 2Stella Maris Scientific Institute and IMAGO7 Research Foundation, Pisa, Italy, 3Dipartimento di Fisica, Università di Pisa, Pisa, Italy, 4Computer Science, Technische Universitat Munchen, Munich, Germany

Magnetic resonance fingerprinting (MRF) is a useful tool for simultaneously obtaining multiple tissue-specific parameters in an efficient imaging experiment. This technique uses transient state acquisitions with pseudo-random acquisition parameters. However, specific schedules may be better suited for certain parameter ranges or sampling patterns. This work aims to introduce a framework for pulse sequence optimization, including aliasing and noise in our estimates, individually or jointly optimizing for T1 and T2 relaxation times. We demonstrated the schedules created by our algorithm using MRI acquisitions on a healthy volunteer. The design framework could improve the efficiency and accuracy of T1 and T2 acquisitions.

4524
Computer 153
Optimization of MR Fingerprinting Sequence Using a Quantum Inspired Algorithm
Dan Ma1, Stephen Jordan2, Rasim Boyacioglu1, Michael Beverland2, Yun Jiang1, Darryl Jacob3, Sherry Huang4, Helmut G Katzgraber2, Julie Love2, Mark A Griswold5, Matthias Troyer2, and Debra F McGivney1

1Radiology, Case Western Reserve University, School of Medicine, Cleveland, OH, United States, 2Microsoft, Seattle, WA, United States, 3Physics and Astronomy, Texas A&M University, College Station, TX, United States, 4Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 5Case Western Reserve University, School of Medicine, Cleveland, OH, United States

MR Fingerprinting (MRF) is a fast quantitative MR imaging technique that simultaneously quantifies multiple tissue properties. We propose to use quantum-inspired optimization to characterize the optimization landscape by using an appropriate cost function to account for signal features and create an optimization frontier. The simulation results from the optimized MRF sequences showed reduced bias and variance as compared to those from the original empirical design. The in vivo maps from the optimized sequences showed improved image quality as well.

4525
Computer 154
An accurate dictionary generation method for MR fingerprinting using a fast Bloch image simulator
Katsumi Kose1 and Ryoichi Kose1

1MRI simulations Inc., Tokyo, Japan

This study proposes a simple and accurate dictionary creation method for MR fingerprinting using a fast Bloch image simulator. A typical MR fingerprinting sequence based on a FISP sequence and a numerical phantom were used for dictionary generation. Cartesian and spiral readout gradients were used for the Bloch image simulation of the numerical phantoms. MR fingerprinting parameter maps obtained by pattern matching with the dictionaries generated by the proposed method demonstrated validity and usefulness of the method. The proposed method is simple and useful for creation of accurate dictionaries in MR fingerprinting.

4526
Computer 155
Towards Continuous Dictionary Resolution in MR Fingerprinting using a Quadratic Inner Product Model
Debra McGivney1, Rasim Boyacioglu1, Yun Jiang1, Charlie Wang2, Dan Ma1, and Mark Griswold1,2

1Radiology, Case Western Reserve University, Cleveland, OH, United States, 2Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States

Magnetic resonance fingerprinting is a framework for creating quantitative tissue property maps from a single acquisition. The accuracy and precision of these maps depend upon a precomputed dictionary of simulated signal evolutions, to which acquired signals are matched using the inner product to determine the tissue property values. We propose to approximate the inner product as a quadratic function of the tissue properties in a neighborhood around the correct match in order to reduce the effect of tissue property step size in the dictionary. Results from data acquired with different MRF sequences demonstrate the value of the proposed approach.

4527
Computer 156
Dictionary-free Reconstruction Based Magnetic Resonance Fingerprinting Optimization
Tianyu Han1, Teresa Nolte1,2, Nicolas Gross-Weege1, and Volkmar Schulz1

1Physics of Molecular Imaging Systems, RWTH Aachen University, Aachen, Germany, 2Multiphysics and Optics, Philips Research Europe, Eindhoven, Netherlands

To make the MRF technique most suitable for clinical needs, efforts are still to be made to accelerate MRF acquisitions while maintaining the accuracy in parameter determination. However, the dictionary calculation is a heavy computational burden for each trial MRF measurement within the optimization process. In this work, we present a numerical study on the optimization of MRF-FISP sequences by using a parallel tempering algorithm. Specifically, an optimization framework tailored for MRF with severe k-space undersampling was developed based on the previously proposed dictionary-free reconstruction (DFR). In vivo measurements were carried out to evaluate the performance of the optimized sequence.

4528
Computer 157
t-Distributed Stochastic Neighbor Embedding (t-SNE) as a Tool for Visualizing the Encoding Capability of Magnetic Resonance Fingerprinting (MRF) Dictionaries
Kirsten Koolstra1, Peter Börnert1,2, Boudewijn Lelieveldt3,4, Andrew Webb1, and Oleh Dzyubachyk3

1C.J. Gorter Center for High Field MRI, Radiology, Leiden University Medical Center, Leiden, Netherlands, 2Philips Research Hamburg, Hamburg, Germany, 3Division of Image Processing, Radiology, Leiden University Medical Center, Leiden, Netherlands, 4Intelligent Systems Department, Delft University of Technology, Delft, Netherlands

In Magnetic Resonance Fingerprinting (MRF), the quality of the parameter maps depends on the encoding capability of the variable flip angle train. In this work we show how the dimensionality reduction technique t-Distributed Stochastic Neighbor Embedding (t-SNE) can be used to obtain insight into the encoding capability of different MRF sequences by embedding high-dimensional MRF dictionaries into a lower-dimensional space and visualizing them as colormaps. Experiments on example dictionaries perform comparison between different sequences and assess the effect of B1+ variations on the encoding capability. 

4529
Computer 158
Optimized fast dictionary matching for magnetic resonance fingerprinting based on echo-planar imaging for enhanced clinical workflow
Ingo Hermann1, Benedikt Rieger1, Jascha Zapp1, Sebastian Weingärtner1,2,3, and Lothar R. Schad1

1Computer Assisted Clinical Medicine, University Medical Center Mannheim, Heidelberg University, Mannheim, Germany, 2Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, United States, 3Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States

In this work, an optimized fast group matching reconstruction for magnetic resonance fingerprinting based on echo-planar imaging was evaluted to enhance clinical usability. This scanner based 'on the fly' reconstruction reduced the reconstruction time by an acceleration factor of 10 shortening the reconstruction to 10 seconds. The fast group matching algorithm was tested in-vivo and compared with full dictionary matching and resulted in virtually no deviation in T1 and T2* maps facilitating the use of MRF in clinical routine.

4530
Computer 159
Magnetic Resonance Fingerprinting Optimization With Variance Based Spiral Arm Ordering
Rasim Boyacioglu1, Debra McGivney1, Dan Ma1, Yun Jiang1, and Mark Griswold1

1Radiology, Case Western Reserve University, Cleveland, OH, United States

Magnetic Resonance Fingerprinting (MRF) maps various tissue properties and system parameters simultaneously. MRF time series, which are matched to a precalculated dictionary, are often obtained with fast acquisition of low resolution images with undersampled spiral trajectories using a regular sampling pattern. In this work, we propose to order a set of spiral trajectories based on dictionary variance instead of the standard sequential or golden-angle ordering. Phantom and in vivo results show that the variance based optimized order converges faster to expected true values. The optimized order does not limit other MRF optimization approaches and can be applied to any MRF sequence.

4531
Computer 160
MR Fingerprinting SChedule Optimization NEtwork (MRF-SCONE)
Ouri Cohen1

1Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, United States

MR Fingerprinting schedule optimization can reduce scan times and improve accuracy but typically relies on minimization of indirect metrics rather than the actual reconstruction error due to the computational challenges involved in calculating the reconstruction error at each iteration of the optimization. Here we introduce a Deep Learning framework that can overcome these challenges and allow direct minimization of the reconstruction error. The proof-of-principle is demonstrated using simulations on a numerical brain phantom.

4532
Computer 161
MRF Dictionary Calculation and Visualization using GPU Compute Shaders
Andrew Dupuis1,2, Dan Ma3, and Mark A Griswold1,2,3

1Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 2Interactive Commons, Case Western Reserve University, Cleveland, OH, United States, 3Department of Radiology, School of Medicine, Case Western Reserve University, Cleveland, OH, United States

Dictionary generation for Magnetic Resonance Fingerprinting (MRF) can be a computationally intensive procedure, especially as complexity and density increase. Conveniently, the majority of operations required for calculating dictionary entries are already enumerated in conventional computer graphics shader packages. Here, we leverage the decades of research and hardware development spent to improve computer graphics optimization to remove the need for CUDA parallelization and instead directly render MRF dictionaries into compressible video files in virtually real time.

4533
Computer 162
Constrained Ellipse Fitting for Efficient T1-T2 Mapping in Phase-cycled bSSFP Imaging
Kübra Keskin1,2 and Tolga Çukur1,2,3

1Electrical and Electronics Engineering, Bilkent University, Ankara, Turkey, 2National Magnetic Resonance Research Center (UMRAM), Bilkent University, Ankara, Turkey, 3Neuroscience Graduate Program, Bilkent University, Ankara, Turkey

There is growing interest in use of balanced steady-state free precession (bSSFP) imaging for simultaneous mapping of $$$T_1$$$, $$$T_2$$$ and off-resonance. An elegant ellipse fitting approach in the complex plane was recently proposed for parameter estimation from multiple phase-cycled acquisitions. Since this approach requires at least six phase-cycles, it can limit scan efficiency. Here, we propose a new technique that integrates a geometric solution with constrained ellipse fitting to enable mapping with only four phase-cycled acquisitions. The proposed method yields accurate $$$T_1$$$, $$$T_2$$$ and off-resonance maps while significantly improving scan efficiency.

4534
Computer 163
Extracting Gold Standard Relaxation Times and Field Map Estimates from the Balanced SSFP Frequency Profile by Neural Network Fitting
Rahel Heule1 and Klaus Scheffler1,2

1High Field Magnetic Resonance, Max Planck Institute for Biological Cybernetics, Tübingen, Germany, 2Department of Biomedical Magnetic Resonance, University of Tübingen, Tübingen, Germany

It has been observed that the balanced steady-state free precession (bSSFP) frequency profile exhibits asymmetries if the intra-voxel frequency content is inhomogeneous and asymmetric. Recent attempts to calculate T1 and T2 values of human brain tissues from the measured bSSFP profile fail to account for anisotropies in the tissue microenvironment and are thus subject to a considerable bias, in particular for white matter. To eliminate this bias, a feedforward neural network is trained with the bSSFP profile as input and a multi-parametric output (i.e., T1, T2, B1, ∆B0) using gold standard relaxation times and reference field maps as ground truth.

4535
Computer 164
Noise Reduction and Uncertainty Estimation for the Variable Flip Angle T1 Method with Automatic Selection of Regularization Parameters
Anders Garpebring1, Max Hellström1, Mikael Bylund1, and Tommy Löfstedt1

1Radiation Sciences, Umeå University, Umeå, Sweden

The purpose of this work was to develop a method that simultaneously reduces and estimates the uncertainty in the T1 maps obtained with the VFA method while also avoiding the need for any manual tuning of regularization parameters. A Markov Chain Monte Carlo-based algorithm was implemented and evaluated on real and synthetic data. The results show that the method can be used to reduce both noise and noise-induced bias and simultaneously give information about the uncertainty in the estimates.

4536
Computer 165
Optimised T2 Preparation for Brain Imaging: Application to Compressed Sensing 3D T2 Mapping
Emilie Mussard1,2,3, Tom Hilbert1,2,3, Christoph Forman4, Ruud B. van Heeswijk2,5, Reto Meuli2, Jean-Philippe Thiran3, and Tobias Kober1,2,3

1Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, 2Department of Radiology, University Hospital (CHUV), Lausanne, Switzerland, 3LTS5, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, 4Magnetic Resonance, Siemens Healthcare GmbH, Erlangen, Germany, 5Center for BioMedical Imaging (CIBM), Lausanne and Geneva, Switzerland

T2-mapping is becoming an important tool to detect pathological tissue; however, achieving high isotropic resolution is challenging. This work optimises a T2-prepared 3D compressed-sensing acquisition. Two T2-preparation modules (modified-BIR4, hyperbolic secant) and three Cartesian sampling trajectories (spiral, radial, VC-spiral) are explored. The NIST-ISMRM phantom and three in vivo subjects were scanned to test T2 accuracy and homogeneity. Results show more homogeneous and accurate T2 values with BIR4, due to a decreased sensitivity to B1. In vivo data showed more homogeneous T2 in WM using a radial trajectory. Based on these results, we propose an optimised 3D T2-mapping protocol of 9:48min.

4537
Computer 166
Practical Considerations for Mapping R1 in the Cerebral Cortex Across Sites
Laagi Yoganathan1, Kim Desmond1, Manpreet Sehmbi2, Benicio N Frey2, and Nicholas Bock1

1Psychology, Neuroscience & Behaviour, McMaster University, Hamilton, ON, Canada, 2Psychiatry & Behavioural Neuroscience, McMaster University, Hamilton, ON, Canada

We map R1 in the cortex across two sites, using IR-GRE and GRE images to calculate R1 values based on the ratio of the images (IR-GRE/GRE)) using signal equations. We collect B1+ maps to analytically correct  R1 inhomogeneities that might cause site-dependent variation. We tested our R1 mapping method with two different input ratio images: one formed using an IR-GRE sequence with typical neuroanatomical contrast, and one using an IR-GRE sequence optimized to produce strong intracortical contrast. We found the ratio image  with the higher intracortical contrast produced more consistent R1 maps across sites, which were less sensitive to B1+.

4538
Computer 167
Sparse MR-STAT: Order of magnitude acceleration in reconstruction times
Oscar van der Heide1, Alessandro Sbrizzi1, Peter R. Luijten1, and Cornelis A.T. van den Berg1

1Center for Image Sciences, UMC Utrecht, Utrecht, Netherlands

MR-STAT is a framework for obtaining multi-parametric quantitative MR maps using data from single short scans. A large-scale optimization problem is solved in which spatial localisation of signal and estimation of tissue parameters are performed simultaneously by directly fitting a Bloch-based volumetric signal model to the time domain data. In the current work, we exploit sparsity that is inherently present in the problem when using Cartesian sampling strategies to achieve an order of magnitude acceleration in reconstruction times. The new method is tested on synthetically generated data and on in-vivo brain data.

4539
Computer 168
Accelerated T2 mapping based on Bloch signal-model with fixed rank and sparsity constraints
Daniel Grzeda1, Meirav Galun2, Noam Omer1, Tamar Blumenfeld-Katzir1, Dvir Radunsky1, Ricardo Otazo3, and Noam Ben-Eliezer1,4,5

1Department of Biomedical Engineering, Tel Aviv University, Tel Aviv, Israel, 2Department of Computer Science and Applied Mathematics, Weizmann Institute of Science, Rehovot, Israel, 3Departments of Medical Physics and Radiology, Memorial Sloan Kettering Cancer Center, New York, NY, United States, 4Sagol School of Neuroscience, Tel Aviv University, Tel Aviv, Israel, 5Center for Advanced Imaging Innovation and Research (CAI2R), New-York University Langone Medical Center, New York, NY, United States

Quantification of T2 values is valuable for a wide range of research applications and clinical pathologies.  Multi-echo spin echo (MESE) protocols offer significantly shorter scan-times, at the cost of strong contamination from stimulated and indirect echoes.  The echo-modulation-curve (EMC) algorithm, can efficiently overcome these limitations to produce accurate T2 values.  In this work we propose a new reconstruction algorithm based on Sparsity and Fixed Rank constraints, denoted as SPARK. We compare our method against GRAPPA and show its superiority in the quantitative evaluation of T2 values from highly undersampled data.


4540
Computer 169
Study of key properties behind a good undersampling pattern for quantitative estimation of tissue parameters
Riwaj Byanju1, Stefan Klein1, Alexandra Cristobal Huerta2, Juan A. Hernandez Tamames2, and Dirk H. J. Poot1

1Departments of Medical Informatics and Radiology, Erasmus MC, Rotterdam, Netherlands, 2Departments of Radiology and Nuclear Medicine, Erasmus MC, Rotterdam, Netherlands

Quantitative MR (qMRI) at present is clinically unfeasible due to long scan time. Jointly performing image reconstruction and parameters estimation is expected to allow increased acceleration. In this work, we investigate properties of undersampling patterns that are most relevant for parameter estimation using a Cramer-Rao-Lower-Bound (CRLB) based metric for such an approach. We compare key properties of undersampling patterns and conclude that one of these properties, namely low discrepancy, is most relevant for achieving time-efficient qMRI.

4541
Computer 170
Accelerated R1 or R2 Mapping with Geometric Relationship Constrained Reconstruction Method
Nadine Luedicke Dispenza1, Gigi Galiana2, Dana C Peters3, Robert Todd Constable4,5, and Hemant D Tagare1,3

1Biomedical Engineering, Yale University, New Haven, CT, United States, 2Diagnostic Radiology, Yale University, New Haven, CT, United States, 3Radiology and Biomedical Imaging, Yale University, New Haven, CT, United States, 4Department of Radiology and Biomedical Imaging, Yale University, New Haven, CT, United States, 5Department of Neurosurgery, Yale University, New Haven, CT, United States

In this work we present a constrained reconstruction method that can produce either an R2- or R1- weighted image series, in tandem with the parameter map, from undersampled data.  The method has been demonstrated in vivo for radial TSE, and radial TSE augmented with nonlinear encoding (O-space), and inversion recovery (IR) datasets. The algorithm iteratively calculates the entire series of T2 or T1 weighted images while enforcing the exponential decay posed as a geometric relationship between the images. Experimental brain images generated with these maps are in excellent agreement with the fully sampled images and show less undersampling artifact than images reconstructed from individual undersampled datasets.

4542
Computer 171
Reconstruction of Tailored Magnetic Resonance Fingerprinting Using Random Forest Approach
Shivaprasad Ashok Chikop1, Amaresh Shridhar Konar1,2, Vineet Vinay Bhombore1, Fabian Balsiger3, Rajagopalan Sundareshan4, shaik Imam4, Mauricio Antonio Reyes Aguirre3, Ramesh venkatesan4, and Sairam Geethanath1,5

1Dayananda Sagar Institutions, Bangalore, India, 2Memorial Sloan Kettering Cancer Center, New York, NY, United States, 3Institute of Surgical Technology and Biomechanics, University of Bern, Bern, Switzerland, 4Wipro-GE, bangalore, India, 5Magnetic Resonance Research Program, Columbia University, New York, NY, United States

Magnetic Resonance Fingerprinting is a new acquisition/reconstruction technique to obtain multi-parametric map. Tailored MRF has demonstrated the quantification of longer T2  components contrary to classical MRF. The supervised learning based approach model in the study does not require construction of the dictionary. Leave out one approach has been utilized as the approach for modeling the random forest approach. The dictionary approach is heavy on the computation that limits the MRF to get into the clinic.

4543
Computer 172
Dynamic streaking artifact regularization for QSM
Yongquan Ye1, Xueping Li2, Qinglei Zhang2, Fei Zhou2, Ming Li2, Zhao Qing2, Bing Zhang2, Shuheng Zhang3, Yanling Chen3, and Jinguang Zong3

1UIH America, Inc., Houston, TX, United States, 2Radiology, The affiliated Drum Tower hospital of Nanjing university medical school, Nanjing, China, 3United Imaging of Healthcare, Shanghai, China

We propose dynamically estimate, formulate and update the field components that are responsible for causing streaking artifact, as an additional regularization term for solving the QSM optimization problem. As a result, streaking artifacts arising from regions with highly disrupted local fields can be well suppressed, preventing them from spatially extending and affecting other regions of interest. The proposed method can maintain the accuracy of QSM results, and has the potential to be integrated into most QSM optimization algorithms.

4544
Computer 173
Sparse Pre-Contrast T1 Mapping for DCE-MRI Calibration
Zhibo Zhu1, R. Marc Lebel2, Yannick Bliesener1, and Krishna S. Nayak1

1Electrical Engineering, University of Southern California, Los Angeles, CA, United States, 2GE Healthcare, Calgary, AB, Canada

Quantitative DCE-MRI requires fast pre-contrast T1 mapping (scan time <3 min) with matching resolution and coverage. Recent advances in imaging have substantially improved resolution and coverage of DCE-MRI but without matched improvements in the pre-contrast T1 data. Here, we demonstrate a sparse T1 mapping method and characterize a tradeoff between data acquisition and  T1 statistics, using a variable flip angle (VFA) approach and sparse Cartesian spiral sampling pattern, with image domain wavelet sparsity constraint. This method provides the necessary high-resolution whole-brain T1/M0 maps for DCE-MRI tracer kinetic analysis.

4545
Computer 174
A fast method for field map calculation in multispectral imaging near metal implants
Yuan Zheng1, Yu Ding1, Yongquan Ye1, and Weiguo Zhang1

1UIH America, Houston, TX, United States

Multispectral acquisition is an important technique for MRI near metal. It is critical to estimate the field map and correct for displacements among bin images before bin combination in order to eliminate blurring. However, current field-estimation methods are either susceptible to noise or are computationally intensive, limiting their clinical applications. We propose a robust and efficient algorithm for calculating the field map from multispectral datasets based on a previous matched-filter field estimation technique. The proposed technique was tested on a digital phantom and generated accurate field maps and high quality images with a very short calculation time.


Quantitative Mapping: Relaxometry & Beyond

Exhibition Hall
Thursday 9:15 - 10:15
 Acquisition, Reconstruction & Analysis

4546
Computer 1
A robust pulse sequence for simultaneous diffusion MRI and MR elastography (diffusion-MRE)
Daiki Ito1,2,3, Tomokazu Numano1,3, Kazuyuki Mizuhara3,4, Toshikatsu Washio3, Masaki Misawa3, Naotaka Nitta3, and Tetsushi Habe1

1Department of Radiological Sciences, Graduate School of Human Health Sciences, Tokyo Metropolitan University, Tokyo, Japan, 2Office of Radiation Technology, Keio University Hospital, Tokyo, Japan, 3Health Research Institute, National Institute of Advanced Industrial Science and Technology, Ibaraki, Japan, 4Department of Mechanical Engineering, Tokyo Denki University, Tokyo, Japan

Diffusion-magnetic resonance elastography (dMRE) can acquire diffusion and mechanical properties simultaneously. However, intravoxel phase dispersion (IVPD) interferes with the calculation of the apparent diffusion coefficient (ADC). This study presents an approach to dMRE that reduces the influence of IVPD by introducing a new pulse sequence. The ADC and stiffness, obtained using the existing and proposed dMRE techniques, were compared with spin-echo (SE)-diffusion and SE-MRE, for a phantom. In existing dMRE technique, the ADC was changed by IVPD but that of proposed dMRE technique was unchanged. The results demonstrate that our dMRE technique is a robust method for addressing the IVPD.

4547
Computer 2
Deep parameter mapping with relaxation signal model driven constraints
Zhiyang Fu1, Sagar Mandava1, Zhitao Li1, Diego R Martin2, Maria I Altbach2, and Ali Bilgin1,2,3

1Electrical and Computer Engineering, University of Arizona, Tucson, AZ, United States, 2Department of Medical Imaging, University of Arizona, Tucson, AZ, United States, 3Biomedical Engineering, University of Arizona, Tucson, AZ, United States

Conventional MR parameter mapping suffers from long acquisition times limiting their clinical utility. Model based iterative methods have been proposed to allow reconstructions from highly accelerated data, but these suffer from high computational costs. Deep learning based methods that can reduce reconstruction times significantly while yielding reconstruction quality comparable to the model based methods have emerged recently. In this work, we evaluate the use of signal model driven constraints in deep learning based MR parameter mapping.

4548
Computer 3
Robust 3D Bloch-Siegert based B1+ mapping using Multi-Echo General Linear Modelling
Nadège Corbin1, Julio Acosta-Cabronero1, Shaihan J Malik2, and Martina F Callaghan1

1Wellcome Centre for Human Neuroimaging, UCL Institute of Neurology, London, United Kingdom, 2School of Biomedical Engineering & Imaging Sciences, King’s College London, London, United Kingdom

Robust quantification of the longitudinal relaxation rate (R1)—a widely used proxy marker of myelin content—requires highly accurate and precise estimation of the RF transmit field (B1+). The Bloch-Siegert shift (BSS) is a B1+-mapping method that allows calibration data to be acquired with the same spoiled gradient-echo readout used for variable flip angle R1 mapping. Here we show that systematic differences in steady state phase, caused by the interleaved nature typically adopted, lead to bias or loss of precision, but that these effects can be corrected for using a multi-echo approach and GLM fitting to isolate the BSS phase.

4549
Computer 4
3-Dimensional Strain Mapping of the Eyeball during Adduction, Abduction Tasks
David J Ouellette1, Kenneth Wengler1, Patrick Sibony2, Xiang He3, and Tim J Duong3

1Biomedical Engineering, Stony Brook University, Stony Brook, NY, United States, 2Ophthalmology, Stony Brook University, Stony Brook, NY, United States, 3Radiology, Stony Brook University, Stony Brook, NY, United States

This study employed high-resolution 3D-MRI to map the strain of the eyeball during adduction and abduction tasks. The strain map is highly heterogeneous with high strain toward the anterior region. Adduction induced higher strain than abduction, as expected due to more stretching of the optic nerve in the adduction position. This is the first MRI measurement of strain of the eyeball. This approach could have clinical applications in eye movement disorders and eye diseases.

4550
Computer 5
High resolution 3D magnetic resonance fingerprinting with hybrid radial cartesian-EPI acquisition
Dongyeob Han1, Taehwa Hong1, and Dong-Hyun Kim1

1Yonsei University, Seoul, Korea, Republic of

A high resolution (0.5x0.5x1mm3) 3D MRF method was proposed using a hybrid radial cartesian-EPI acquisition with both segmented & interleaved EPI strategy. For the reconstruction, k-space SVD compression and CG-SENSE were applied. An in vivo brain results were presented.

4551
Computer 6
Assessment of Absolute pH Using and Magnetic Resonance Fingerprinting and a Single Dysprosium-Based MRI Contrast Agent
Yifan Zhang1, Andre F Martins2, Dean A Sherry2, Christian E Anderson1, and Chris A Flask1

1Radiology, Case Western Reserve University, Cleveland, OH, United States, 2Chemistry and Biochemistry, The University of Texas at Dallas, Richardson, TX, United States

In this initial in vitro study, we used Magnetic Resonance Fingerprinting (MRF)-based T1 and T2 relaxation time maps to estimate the linear relationship between pH and relaxivity (r1 and r2) for a previously-described dysprosium (Dy) MRI contrast agent. These relaxivity estimates were then used to calculate MRF-based estimates of pH for each solution for comparison with gold-standard measurements by pH electrode at 7.0T (R = 0.93, p = <1e-6) and 9.4T (R = 0.68, p = 0.004).  Results show MRF can be used in combination with a pH-sensitive paramagnetic MRI contrast agent to accurately estimate pH  independent of agent concentration. 

4552
Computer 7
Magnetic Resonance Fingerprinting with Pure Quadratic RF Phase
Charlie Yi Wang1, Rasim Boyacioglu2, Debra McGivney2, Dan Ma2, Xin Yu1,2, and Mark Griswold2

1Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 2Radiology, Case Western Reserve University, Cleveland, OH, United States

Previous work has shown that Magnetic Resonance Fingerprinting with quadratic RF phase (qRF-MRF) can be used to simultaneously quantify off-resonance, T1, T2 and T2*.  This method employed a mix of bSSFP and qRF pulse sequence block segments for reliable tissue property quantification.  However, the incorporation of bSSFP type acquisition schemes resulted in null-band artifacts near bSSFP signal voids.  Here, we present a bSSFP-free pure qRF-MRF method with elimination of null-band artifacts, and explore its potential for tissue property mapping with reduced acquisition time.

4553
Computer 8
Combination of ESPIRiT and back-projection reconstruction for 3D MR fingerprinting within 2.5 minutes
Xiaozhi Cao1, Qing Li1, Huihui Ye1,2, Hongjian He1, and Jianhui Zhong1,3

1Center for Brain Imaging Science and Technology, Department of Biomedical Engineering, Key Laboratory for Biomedical Engineering of Ministry of Education, Zhejiang University, Hangzhou, China, 2State Key Laboratory of Modern Optical Instrumentation, College of Optical Science and Engineering, Zhejiang University, Hangzhou, China, 3Department of Imaging Sciences, University of Rochester, Rochester, NY, United States

A spiral projection acquisition scheme was implemented for 3D MR fingerprinting to achieve isotropic resolution of 1x1x1 mm3 in whole brain T1 and T2 mapping within 2.5 minutes by using efficient L1SPIRiT reconstruction (ESPIRiT) and back-projection reconstruction. 

4554
Computer 9
Rapid, simultaneous non-synthetic multi-contrast and quantitative imaging using Tailored MR Fingerprinting (TMRF)
Sairam Geethanath1, Sachin Jambawalikar1,2, Maggie Fung3, Angela Lingella2, and John Thomas Vaughan1

1MR Research Center, Columbia University, New York, NY, United States, 2Radiology, Columbia University, New York, NY, United States, 3GE Healthcare, New York, NY, United States

The goal of this work was to rapidly acquire non-synthetic multi-contrast and quantitative images simultaneously, through tailoring the MR fingerprinting acquisition schedule in contrast blocks. TMRF providing for five contrasts was designed, simulated and demonstrated on four healthy volunteer brain scans. The acquisition times for MRF and TMRF were 5:11 and 4:41 (min: sec) respectively.  The spatio-temporal profiles of T1, T2, PD, water-fat and flow contrasts were reconstructed block-wise along with relaxometric maps. Comparatively, TMRF images showed higher mean to standard deviation ratios for the four volunteers over the contrast blocks for PD and T2 while maintaining similarity of relaxometric maps.

4555
Computer 10
EPI based Dual-stage MR Fingerprinting for T1, T2, and T2* mapping
Young-Jung Yang1, Pan Ki Kim1,2,3, Dong Jin Im1, Donghyun Hong4, and Byoung Wook Choi1

1Department of Radiology and Research Institute of Radiological Science, Severance Hospital, Yonsei University College of Medicine, Seoul, Korea, Republic of, 2Center for NanoMedicine, Institute for Basic Science (IBS), Seoul, Korea, Republic of, 3Yonsei-IBS Institute, Yonsei University, Seoul, Korea, Republic of, 4Erwin L. Hahn Institute for Magnetic Resonance Imaging, University of Duisburg-Essen, Essen, Germany

We propose an improved MR fingerprinting which can generate T1, T2, and T2* maps simultaneously. This method is based on single-shot EPI and signal acquisition consists of dual-stage divided by fixed and variable echo time. Dictionary generation and pattern matching were also modified in accordance with acquisition scheme. The feasibility of proposed method was demonstrated by phantom study and the MRF results are well correlated with the conventional T1, T2, and T2* maps. In-vivo brain MRF was also performed with a healthy volunteer.

4556
Computer 11
GFB-MRF: Parallel spatial and Bloch manifold regularized iterative reconstruction for magnetic resonance fingerprinting
Simon Arberet1, Xiao Chen1, Boris Mailhe1, Peter Speier2, Mathias Nittka2, Heiko Meyer2, and Mariappan Nadar1

1Digital Services, Digital Technology & Innovation, Siemens Medical Solutions, Princeton, NJ, United States, 2Siemens Healthcare, Application Development, Erlangen, Germany

We introduce a new iterative algorithm for Magnetic Resonance Fingerprinting (MRF) where spatial regularization and fingerprint matching are applied in parallel. This enables to have simultaneously a spatial regularization in addition to the time domain Bloch manifold regularization. Our proposed algorithm showed significant improvements with respect to the state of the art in particular regarding the robustness with respect to measurement noise.

4557
Computer 12
Accelerated Multi-band Magnetic Resonance Fingerprinting Using Spiral in-out with additional kz Encoding and Modified Sliding Window Reconstruction
Di Cui1, Xiaoxi Liu1, Hing-Chiu Chang1, Queenie Chan2, and Edward S Hui1

1Diagnostic Radiology, The University of Hong Kong, Hong Kong, China, 2Philips Healthcare, Hong Kong, China

Multi-band Magnetic Resonance Fingerprinting can be achieved using UNFOLD-like acquisition and dictionary matching without using parallel methods. However, the MR parametric maps after dictionary matching in one slice suffers from artifacts due to the high frequency components of other simultaneously acquired slices. In this work, a new acquisition strategy was proposed for the multi-band acquisition, where spiral-in-out trajectory was used to provide extra kz encoding. A modified sliding window reconstruction was also proposed to reduce the high frequency oscillations.

4558
Computer 13
Computational method for T2-weighted images based on polynomial approximation using 3D MR parameter mapping with RF-spoiled gradient echo
Tomoki Amemiya1, Suguru Yokosawa1, Yo Taniguchi1, Ryota Sato1, Hisaaki Ochi1, and Toru Shirai1

1Research & Development Group, Hitachi, Ltd., Tokyo, Japan

We propose a computational method for obtaining T2-weighted images from maps of proton density, T1, and T2* acquired by 3D RF-spoiled gradient echo. The proposed method uses a predetermined polynomial that approximates the relationship between the MR parameters and the intensity of T2WI on the basis of datasets of other subjects. Similarities between computed images and actually scanned images were improved compared with a computation method using T2* instead of T2 in the theoretical equation of the spin echo signal.

4559
Computer 14
T1 measurement in short-T2 material with suppressed long-T2 component using an IR-UTE multishot sequence
Lucas Soustelle1,2, Julien Lamy2, François Rousseau3, Jean-Paul Armspach2, and Paulo Loureiro de Sousa2

1Aix-Marseille Univ. CRMBM UMR 7339, Marseille, France, 2Université de Strasbourg, CNRS, ICube, FMTS, Strasbourg, France, 3Institut Mines Télécom Atlantique, INSERM, LaTIM, Brest, France

T1 quantification of short-T2 species is challenging due to the uncommon behavior of the signal decay and magnetization tilting during excitations in conventional sequences. UTE sequences can therefore be considered with refined magnetization evolution models using the Bloch equations. In voxels comprising a mix of long and short-T2 components (e.g. myelin and water in the normal appearing white matter), an appropriate long-T2 suppression scheme is mandatory. In this work, we propose an analytical model to quantify the T1 of a short-relaxing component in an accelerated Inversion-Recovery UTE in vitro, and within long-T2 suppression condition.

4560
Computer 15
3D Radial Phase Encoded Flip Angle Imaging at Ultra-High Field Strength
Sebastian Dietrich1, Christoph Kolbitsch1,2, Tobias Schaeffter1,2, and Sebastian Schmitter1

1Physikalisch-Technische Bundesanstalt (PTB), Braunschweig and Berlin, Germany, 2Division of Imaging Sciences and Biomedical Engineering, King's College London, London, United Kingdom

In this work a novel 3D flip angle (FA) mapping method is introduced which combines the advantages of the AFI pulse sequence (3D, low SAR) with the motion robustness of a radial phase encode acquisition scheme. Mapping the FA is needed for human body imaging at 7 tesla to improve image quality and assessment of quantitative results. Therefore, we combined an interleaved acquisition of two FIDs, S1 and S2, acquired for different repetition times TR1 and TR2 with radial phase encode trajectory. In this study we validated this new sequence with a body phantom and in two in-vivo scans. Similar results to cartesian reference scans were obtained and reasonable motion resolved abdominal FA maps were acquired.

4561
Computer 16
Combining Parallel Imaging and Model-based Reconstruction for Isotropic 3D T2 mapping with Multi-Echo GRASE
Gian Franco Piredda1,2,3, Tom Hilbert1,2,3, Reto Meuli2, Josef Pfeuffer4, Jean-Philippe Thiran2,3, and Tobias Kober1,2,3

1Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, 2Department of Radiology, Centre Hospitalier Universitaire Vaudois (CHUV), Lausanne, Switzerland, 3LTS5, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, 4Application Development, Siemens Healthcare GmbH, Erlangen, Germany

Classical T2 mapping based on 2D multi-echo spin-echo sequences achieves only limited across-slice resolutions. For clinical use, acquisitions with high isotropic resolution are however desirable, resulting in clinically prohibitive scan times. To this end, we propose a 3D multi-echo gradient and spin echo sequence with CAIPIRINHA and an additional model-based acceleration for T2 estimation. The combination of these techniques allows for whole-brain T2 mapping with 1.6mm-isotropic resolution in 3:26 min. The proposed framework was tested both in phantom and in vivo experiments.

4562
Computer 17
Quantitative and synthetic MRI using a Multi-Pathway Multi-Echo (MPME) acquisition followed by machine-learning contrast translation
Cheng-Chieh Cheng1, Frank Preiswerk1, and Bruno Madore1

1Radiology, Brigham and Women's Hospital, Harvard Medical School, Boston, MA, United States

Brain exams would ideally include 3D quantitative maps of several basic MR parameters, such as T1, T2, T2* and B0, along with popular qualitative contrasts such as MPRAGE and FLAIR, for example. A multi-pathway multi-echo (MPME) pulse sequence was developed that captured vast amounts of information about the imaged object relatively fast, but not necessarily with image contrasts that radiologists might be comfortable reading. A neural network was trained to act as a ‘contrast translator’, to convert information rapidly obtained from MPME scans into useful quantitative and qualitative contrasts, in effect condensing a whole exam into a single 3D scan.

4563
Computer 18
Variable Flip Angle T1-Mapping Using Perfect In-Phase ZTE
Mathias Engström1, Axel Hartwig1, Cristina Cozzini2, Graeme C McKinnon3, and Florian Wiesinger2

1GE Healthcare, Stockholm, Sweden, 2GE Healthcare, Munich, Germany, 3GE Healthcare, Waukusha, WI, United States

This work details an extension of the Perfect In-Phase ZTE (pipZTE) method that allows for fast and efficient T1-mapping. By adding a variable flip angle scheme in combination with the Perfect In-Phase ZTE readout band-width modulation PD and T1 mapping can be achieved without interference from chemical shift artifacts.  

4564
Computer 19
A Fast Multi-slice T1 mapping method based on SPatiotemporal ENcoding
Qingjia Bao1, Ricardo P. Martinho1, and Lucio Frydman1

1Weizmann Institute of Science, Rehovot, Israel

  A pulse sequence for T1 relaxation time mapping which enables high-resolution and multi-slice imaging in short acquisition times, is presented. The sequence combines fast, robust acquisitions based on SPatiotemporal ENcoding (SPEN), with an accelerated sampling of the T1 inversion curve via slice shuffling. The experiment allows for T1 quantifications in challenging regions subject to field inhomogeneities. In phantoms it provides reliable T1 maps that agree with standard measurements; in in vivo studies of murine brain and kidney models, it provides fewer distortions than comparable EPI-based counterparts.

4565
Computer 20
Accelerating Bi-exponential T1ρ mapping using SCOPE
Yuanyuan Liu1, Yanjie Zhu1, Jing Cheng1, Weitian Chen2, Xin Liu1, and Dong Liang1,3

1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, 2Department of Imaging and Interventional Radiology, The Chinese University of Hong Kong, Shatin, Hong Kong, China, 3Research center for Medical AI, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

Mono-exponential T mapping requires 4 or 5 T-weighted images with different spin lock times (TSLs) to obtain the T  maps, while bi-exponential T mapping requires a larger number of TSLs, which further prolongs the acquisition time. In this work, we develop a variable acceleration rate undersampling strategy to reduce the total scan time. A signal compensation strategy with low-rank plus sparse model was used to reconstruct the T-weighted images. We provide the reconstructed images and the estimated T maps at an acceleration factor up to 6.1 in fast bi-exponential T mapping.

4566
Computer 21
Dual Contrast Weighting and Simultaneous T2 and T2* Mapping with Radially Sampled RARE-EPI
Carl Julius Jacob Herrmann1, Katharina Paul1, Till Huelnhagen1, and Thoralf Niendorf1

1Berlin Ultrahigh Field Facility (B.U.F.F.), Max Delbrueck Center for Molecular Medicine in the Helmholtz Association, Berlin, Germany

MRI examinations commonly involve a series of multiple imaging contrasts and MR-metrics. Dual or even more contrast techniques offer substantial scan time reduction and eliminate the propensity to slice co-registration errors induced by bulk and physiological motion. Recognizing this opportunity this work presents a dual contrast RARE-EPI hybrid, that provides T2 (RARE module) and T2* (EPI module) contrast and facilitates simultaneous T2 and T2* mapping in a single radially (under)sampled scan (2-in-1 RARE-EPI). The applicability of 2-in-1 RARE-EPI is demonstrated in phantom and in in vivo studies and benchmarked versus conventional T2 and T2* weighted/mapping techniques.

4567
Computer 22
A novel strategy to perform the dual flip angle method for the fast and accurate T1 mapping by MRI
Steven Yee1, Michael Fadell2, Mark S Brown2, and Seonghwan Yee2

1SUNY Geneseo, Geneseo, NY, United States, 2University of Colorado School of Medicine, Aurora, CO, United States

Fast T1 mapping can be done by utilizing dual flip angles in acquiring spoiled gradient echo signals. However, its accuracy may be questionable even when the suggested optimal flip angle pair is used. Noting that the faithful action of the prescribed flip angles is the key to the accuracy, we present here a novel dual flip angle method by which the system-specific RF-pulse fidelity of flip angles can be validated and, if necessary, calibrated to improve the T1 accuracy in a wide in vivo range. We tested this method on a few 1.5 or 3T MRI systems of major vendors.

4568
Computer 23
Evaluation of different colormaps for best visual assessment of quantitative Magnetic Resonance Fingerprinting data
Verena Carola Obmann1,2, Ananya Panda3, Chaitra Badve1,4, Jeffrey Sunshine1,4, Vikas Gulani1,4, and Mark Griswold1,4

1Radiology, Case Western Reserve University, Cleveland, OH, United States, 2Diagnostic, Pediatric and Interventional Radiology, Inselspital, Bern, Switzerland, 3Radiology, Mayo Clinic, Rochester, MN, United States, 4Radiology, University Hospitals Cleveland Medical Center, Cleveland, OH, United States

Increasingly quantitative methods such as apparent diffusion coefficient, T1, T2 and T2* mapping or elastography are used in MR imaging. As quantitative data provide multidimensional characterization of pathophysiology, color provides an additional dimensionality to visualize the data. This study demonstrates the superiority of three different colormaps over grayscale display of each T1 and T2 maps for MR Fingerprinting.

4569
Computer 24
Relaxation parameter estimation from limited time points
Tianle Zhao1, Weitian Chen2, and Thierry Blu1

1Department of Electronic Engineering, The Chinese University of Hong Kong, Shatin, Hong Kong, 2Department of Imaging and Interventional Radiology, The Chinese University of Hong Kong, Shatin, Hong Kong

T1rho is a useful biomarker for the diagnosis of several diseases. Current imaging techniques usually use uniform sampling and require a relatively large number of samples to get reliable estimations of T1rho. We show that the intuitive uniform sampling is not optimal, and propose an optimal sampling strategy. We also propose a fast estimation algorithm, which (with the use of spatial redundancy) provides accurate estimates of the T1rho relaxation map from as few as 3 different spin-lock time samples.

4570
Computer 25
Predicting pathological subtypes and stages of thymic epithelial tumors using DWI: value of combining ADC and texture parameters
Bo Li1, Xiao-cheng Wei2, Yu-chuan Hu1, and Guang-bin Cui1

1Tangdu Hospital, Department of Radiology,Fourth Military Medical University, Xi’an, China, 2GE Healthcare China, Xi'an, China

To explore the value of combining apparent diffusion coefficients (ADC) and texture parameters from diffusion-weighted imaging (DWI) in predicting the pathological subtypes and stages of thymic epithelial tumors (TETs). In this study, Fifty-seven patients with TETs confirmed by pathological analysis were retrospectively enrolled. The results showed combination of ADC and DWI texture parameters improved the differentiating ability of TET grades, which could potentially be useful in clinical practice regarding the TETs evaluation before treatment.


Going Faster: New Sequences & Acquisition Protocols

Exhibition Hall
Thursday 9:15 - 10:15
 Acquisition, Reconstruction & Analysis

4571
Computer 26
Rapid, multi-TE, T2-prepared RUFIS for Silent T2-weighted imaging
Emil Ljungberg1,2, Brian Burns3, Tobias Wood1, Shannon Kolind4, Florian Wiesinger5, and Gareth J Barker1

1Neuroimaging, King's College London, London, United Kingdom, 2General Electric Healthcare, London, United Kingdom, 3ASL West, General Electric Healthcare, Menlo Park, CA, United States, 4Medicine, University of British Columbia, Vancouver, BC, Canada, 5ASL Europe, General Electric Healthcare, Munich, Germany

Zero echo time (ZTE) imaging using the RUFIS sequence allows for silent imaging with high efficiency. Without modifications, RUFIS produces proton density and/or T1-weighted images similar to a spoiled gradient echo sequence. In this work we present a novel T2-prepared RUFIS sequence with multiple echo times acquired in each shot, for efficient T2-weighted imaging. We present in vivo results acquired in 11 min with 1.5mm3 resolution, with effective echo times from 0 to 248ms.

4572
Computer 27
Rapid High Resolution T1-Weighted Hippocampus Imaging with Yarn-Ball Acquisition
Rob Stobbe1, Peter Seres1, Don Gross1, and Christian Beaulieu1

1University of Alberta, Edmonton, AB, Canada

For spoiled steady-state T1-weighted imaging, readout duration (TRO) and repetition time (TR) increase result in greater contrast-to-noise ratio (CNR) efficiency. Novel 3D-twisting Yarnball acquisition realizes this advantage without scan-time penalty (more of k-space sampled with increased TRO), but increased TRO results in greater point-spread-function smearing. Following TRO optimization, Yarnball is used to produce whole-brain 0.36x0.36x1.08 mm3 coronal (defined by 1/(2kmax)) images in 10 minutes (with 2 averages). Compared to 3D-MP-RAGE (same scan time and voxel volume) Yarnball images have greater resolution and grey-white CNR, facilitating sharper depiction of internal hippocampus architecture.    

4573
Computer 28
3D SPARKLING for accelerated ex vivo T2*-weighted MRI with compressed sensing
Carole Lazarus1,2, Pierre Weiss3,4,5, Franck Mauconduit6, Alexandre Vignaud1, and Philippe Ciuciu1,2

1NeuroSpin, CEA Saclay, Gif-sur-Yvette, France, 2Parietal, INRIA, Saclay, France, 3ITAV, Toulouse, France, 4CNRS, Toulouse, France, 5Université de Toulouse, Toulouse, France, 6Siemens Healthineers, Saint-Denis, France

In the last decade, compressed sensing (CS) has been successfully used in MRI to reduce the acquisition time. Recently, we have proposed a new optimization-driven algorithm to design optimal non-Cartesian sampling patterns for CSMRI, called SPARKLING for Spreading Projection Algorithm for Rapid K-space sampLING. This method has a few advantages compared to standard trajectories such as radial lines or spirals: i) it allows to reproduce arbitrary densities while the other two are restricted to radial densities and ii) it is more robust to system imperfections. In this communication, we introduce an extension of the SPARKLING method for 3D imaging that allows to achieve an isotropic resolution of 600 μm in just 45 seconds for T2*-weighted ex vivo brain imaging at 7 Tesla.

4574
Computer 29
Capturing Time-Dependent Electric Currents Using MRI with A Sub-Millisecond Temporal Resolution
Zheng Zhong1,2, Muge Karaman1,2, Theodore Claiborne1, and Xiaohong Joe Zhou1,2,3,4

1Center for MR Research, University of Illinois at Chicago, Chicago, IL, United States, 2Bioengineering, University of Illinois at Chicago, Chicago, IL, United States, 3Radiology, University of Illinois at Chicago, Chicago, IL, United States, 4Neurosurgery, University of Illinois at Chicago, Chicago, IL, United States

Neuronal current mapping using MRI has profound biomedical applications, but is hampered by limited temporal resolution. Using a technique known as Sub-Millisecond Imaging of cycLic Event or SMILE, we demonstrate that the temporal resolution of MRI can be substantially increased to the sub-millisecond scale or shorter. This allows capturing ultra-fast physical or biological processes that are cyclic. Although our experimental studies are limited to mapping time-varying currents in a phantom, the same concept can be extended to capturing more complex biological processes, including but not limited to, neuronal currents.

4575
Computer 30
Spin-Echo ZTE-BURST for Quiet T2-Weighted Imaging
Rolf F Schulte1 and Ana Beatriz Solana1

1GE Healthcare, Munich, Germany

ZTE was combined with single spin-echo BURST encoding for acquiring T2 weighted images in a relatively quiet manner.

4576
Computer 31
Silent T2* Imaging on 7T using ZTE Combined with Gradient-Echo BURST
Rolf F Schulte1, Mauro Costagli2, Ana Beatriz Solana1, and Guido Buonincontri2

1GE Healthcare, Munich, Germany, 2IRCCS Stella Maris Foundation and IMAGO7, Pisa, Italy

ZTE combined with gradient-echo BURST enables silent 3D radial T2* imaging. It was implemented on 7T and T2* weighted images were acquired with isotropic resolutions of 1-3mm. From the series with different echo times, both phase and T2* maps were extracted.

4577
Computer 32
Variable Frequency Wave-encoded 3D Turbo Spin Echo Imaging
Zechen Zhou1, Baocheng Chu2, Chun Yuan2,3, and Peter Börnert4

1Philips Research North America, Cambridge, MA, United States, 2Vascular Imaging Lab, Department of Radiology, University of Washington, Seattle, WA, United States, 3Center for Biomedical Imaging Research, Department of Biomedical Engineering, Tsinghua University, Beijing, China, 4Philips Research Hamburg, Hamburg, Germany

Wave encoding is an emerging approach that can take better usage of the three-dimensional (3D) spatial encoding power of multi-channel coils employed in parallel imaging (PI). In this work, a variable frequency (VF) wave encoding approach is proposed to improve the aliasing propagation property and reduce the side lobe amplitude of the transformed point spread function. This VF approach can also induce amplitude modulated wave encoding gradients to reduce eddy currents and improve the slice selection profile. The preliminary results demonstrated its improved PI performance for 3D turbo spin echo imaging over Cartesian and constant frequency wave encoding schemes.

4578
Computer 33
Accelerating Three-Dimension Balanced Steady-State Free Precession Imaging with Modified Wave-CAIPI Technique
Shi Su1, Dong Liang1,2, Zhilang Qiu1, Caiyun Shi1, Xin Liu1, Hairong Zheng1, and Haifeng Wang1

1Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, 2Research Center for Medical AI, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

Balanced steady-state free precession (bSSFP) has merits such as high signal-to-noise ratio, T2/T1 contrast and rapid acquisition speed. However, bSSFP requires further acceleration in 3D imaging due to massive data collected. The acceleration of conventional parallel imaging techniques is limited. In this study, we propose wave-bSSFP by using a modified wave-CAIPI technique to highly accelerate bSSFP. Wave gradients were truncated to further reduce g-factor noise penalty with high wave amplitudes. The simulation and in vivo experiment indicate that wave-bSSFP is effective in decreasing g-factor. Here, an acceleration factor of 9 was achieved in brain scan with 0.8 mm isotropic resolution.

4579
Computer 34
Highly Accelerated 3D EPI using Compressed Sensing
Patrick Liebig1,2,3, Robin Martin Heidemann2, Bernhard Hensel1, and David Andrew Porter3

1University of Erlangen-Nuremberg, Erlangen, Germany, 2Siemens Healthcare GmbH, Erlangen, Germany, 3University of Glasgow, Glasgow, United Kingdom

In previous work, Echo-Planar Imaging (EPI) has been used in combination with a CAIPIRINHA undersampling scheme, as in SMS blipped CAIPI or 3D CAIPI EPI, for highly accelerated BOLD, perfusion and diffusion weighted imaging. In a separate development, Compressed Sensing (CS) was employed in combination with parallel imaging to significantly accelerate a range of non-EPI 3D imaging sequences. In general, this is achieved by using a variable-density randomized sampling scheme which gives aliasing artefacts a noise like appearance. This work explores the use of CS to accelerate 3D EPI acquisitions and demonstrates an improved performance compared to the CAIPIRINHA approach.

4580
Computer 35
Highspeed Imaging of the Vocal Folds Oscillations with Image-based Motion Correction
Johannes Fischer1, Ali Caglar Özen1,2, Matthias Echternach3, Bernhard Richter4, and Michael Bock1

1Department of Radiology, Medical Physics, Medical Center University of Freiburg, Faculty of Medicine, University of Freiburg, Freiburg, Germany, 2German Consortium for Translational Cancer Research Freiburg Site, German Cancer Research Center (DKFZ), Heidelberg, Germany, 3Division of Phoniatrics and Pediatric Audiology, Department of Otorhinolaryngology, Head and Neck Surgery, Ludwig-Maximilians-University, Munich, Germany, 4Institute of Musicians' Medicine, Freiburg University Medical Center, Germany Faculty of Medicine, University of Freiburg, Freiburg, Germany

Highspeed imaging of the vocal folds oscillations is possible by applying a very short phase encoding gradient along the direction of motion. Due to repeated breathing cycles of the volunteer, motion and shifts are introduced that impair image quality. With the use of phase only cross correlation, we correct for this motion prior to the gated reconstruction by applying a linear phase to the k-space data. The proposed method is shown to improve reconstruction of anatomical features and SNR.

4581
Computer 36
Twisted radial echo planar trajectory (EPIstar) for 3D self-navigated golden angle structural and functional MRI
Christoph Rettenmeier1, Danilo Maziero1, and V. Andrew Stenger1

1Medicine, University of Hawaii, Honolulu, HI, United States

A new 3D trajectory design for efficient, self-navigated golden angle high-resolution MRI acquisition is presented along with results in SWI and BOLD functional MRI.

4582
Computer 37
Simultaneous T2* and T2 weighted imaging based on ultrafast SPEN MRI
Qingjia Bao1 and Lucio Fydman2

1Weizmann Institute of Science, Rehovot, Israel, 2Weizmann Institute of Science, Rehovot, AB, Israel

This work presents new sequences to acquire multislice images with different contrasts –a T2* weighted one for enhancing BOLD and a T2 weighted one for faithful location– in a single shot. The sequences rely on SPatiotemporal ENcoding (SPEN), an ultrafast MRI method with immunity to artifacts, and they utilize a “full-refocusing” mode to obtain a T2 weighted image and a “non full-refocusing” mode to obtain T2* weighted images. Two variants are proposed, differing in the order with which they collect their T2*- and the T2-weighted information, and are demonstrated with in vivo studies of brain and kidney in mice.

4583
Computer 38
On the signal strength of Simultaneous Transmission and Reception (STAR) acquisition: EPG simulation and analysis
Gehua Tong1, Sairam Geethanath2, and John Thomas Vaughan2

1Biomedical Engineering, Columbia University, New York, NY, United States, 2MR Research Center, Columbia University, New York, NY, United States

Simultaneous Transmission And Reception (STAR) has the potential to remove the constraint of temporal separation between transmission and reception. In principle, much shorter acquisition times with significantly higher signal strength compared to pulsed sequences should be achievable. However, the signal characteristics differ from that of the conventional pulsed-RF framework. In this work, we characterize STAR characteristics with extended phase graph (EPG) simulation. We show the signal evolution from a simple STAR experiment as well as how tissue contrast could be generated in steady state.

 


4584
Computer 39
Accelerated Volumetric FRONSAC with WAVE and CAIPI
Nadine Luedicke Dispenza1, Robert Todd Constable2,3, and Gigi Galiana4

1Biomedical Engineering, Yale University, New Haven, CT, United States, 2Department of Radiology and Biomedical Imaging, Yale University, New Haven, CT, United States, 3Department of Neurosurgery, Yale University, New Haven, CT, United States, 4Radiology and Biomedical Imaging, Yale University, New Haven, CT, United States

This work demonstrates the potential of FRONSAC, which adds oscillating nonlinear gradients to the Cartesian readout, for 3D accelerated imaging. In undersampled trajectories using either standard Cartesian encoding, CAIPI encoding, or WAVE-CAIPI encoding, significant further improvements are achieved when FRONSAC is applied in addition to these approaches. 

4585
Computer 40
Wave-CAIPI accelerated whole brain structure imaging using three-dimensional T1 weighted SPACE sequence
Zhilang Qiu1, Sen Jia1, Haifeng Wang1, Xin Liu1, Hairong Zheng1, and Dong Liang1,2

1Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, 2Research Center for Medical AI, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

Three-dimensional (3D) SPACE (sampling perfection with application optimized contrast using different flip angle evolutions) sequences are the workhorse for volume imaging with isotropic spatial resolution. However, spatial resolution is often scarified to achieve clinically acceptable scan time. Conventional one- and two-dimensional parallel imaging techniques could help reducing the scan time but would lead to deteriorated signal-to-noise (SNR) performance at submillimeter spatial resolutions. In this study, three-dimensional parallel imaging technique-Wave-CAIPI is utilized to improve the SNR performance for whole brain SPACE imaging with isotropic 0.6 mm resolution. In vivo results demonstrated that Wave-CAIPI could improve the SNR at 5x acceleration.

4586
Computer 41
Supersonic imaging with a silent gradient axis driven at 20 kHz
Edwin Versteeg1, Dennis Klomp1, Jeroen Hendrikse1, and Jeroen Siero1,2

1University Medical Centre Utrecht, Utrecht, Netherlands, 2Spinoza Center for Neuroimaging, Amsterdam, Netherlands

Gradient inserts allow for faster switching and higher gradient strengths than conventional whole-body gradient coils.  However, the higher gradient performance is accompanied by an increase in acoustic sound pressure. We present a gradient insert that switches at 20 kHz (above human hearing perception) and therefore allows for imaging with an inaudible gradient axis.  Additionally, we introduce a readout scheme for imaging at 20 kHz, and show the first imaging results on a phantom and a healthy volunteer using an inaudible gradient axis.

4587
Computer 42
2D k-space waves for silent EPI acquisitions
Jenni Schulz1, Riccardo Metere1, Dennis Klomp2,3, and David G Norris1,4

1Donders Institute for Brain, Cognition and Behaviour, Radboud University Nijmegen, Nijmegen, Netherlands, 2Department of Radiology, University Medical Center Utrecht, Utrecht, Netherlands, 3MR Coils BV, Zaltbommel, Netherlands, 4Erwin L. Hahn Institute for Magnetic Resonance Imaging, University of Duisburg-Essen, Essen, Germany

High acoustic noise levels in fMRI-acquisitions are not only problematic in terms of undesired activation patterns in the brain, but also with respect to patient comfort. A 2D-EPI sequence is presented which is capable of acquiring fMRI-data in silent mode by using a head insert z-gradient coil. For silent data acquisition, a wave-like k-space trajectory is then required. The sequence is compared to a standard FLASH and EPI acquisition. The measured acoustic noise of the silent 2D-EPI is in the order of the idle mode of the scanner and arises from the sound of the continuously active helium pump.

4588
Computer 43
Self-retraced Spiral In-Out 3D Turbo Spin-Echo Imaging
John P. Mugler III1, Craig H. Meyer2, Thomas Benkert3, Josef Pfeuffer3, and Berthold Kiefer3

1Radiology & Medical Imaging, University of Virginia, Charlottesville, VA, United States, 2Biomedical Engineering, University of Virginia, Charlottesville, VA, United States, 3Application Development, Siemens Healthcare, Erlangen, Germany

The purpose of this work was to perform a preliminary evaluation of a self-retraced spiral in-out trajectory for 3D turbo/fast spin-echo imaging.  By sampling k-space locations twice with a single spiral in-out trajectory, off-resonance effects are robustly attenuated and image quality is improved compared to using a standard spiral in-out trajectory.

4589
Computer 44
Single Breath-Hold Diffusion MRI utilizing a Spiral TSE with Variable Flip Angle Refocusing
Naoharu Kobayashi1 and Michael Garwood1

1Center for Magnetic Resonance Research, Department of Radiology, University of Minnesota, Minneapolis, MN, United States

A single breath-hold diffusion MRI sequence utilizing turbo spin echo (TSE) with variable flip angle refocusing and spiral readout is introduced. Flip angles of the refocus RF pulses were determined with the prospective extended phase graph method to minimize the impact of fluctuating refocusing echo signals in TSE. Spiral k-space sampling made the sequence tolerant to motion. The feasibility of the proposed sequence was tested in in vivo brain and thoracic imaging. The proposed single breath-hold diffusion sequence achieved diffusion-weighted imaging of the thoracic region without clear cardiac motion artifacts.

4590
Computer 45
Reduced Noise Enhancement in Whole-Brain Multiband-RASER
Ute Goerke1

1Radiology, CMRR/University of Minnesota, Minneapolis, MN, United States

High acceleration is an effective tool to achieve whole brain coverage with acceptable total acquisition times. A limitation of parallel imaging is high noise amplification at high acceleration factors. In this work, an algorithm for parallel imaging of RASER in two dimensions implemented. The noise enhancement is theoretically derived and experimentally validated. Results show that even at high acceleration noise amplification remains low and high resolution whole brain images can be obtained with RASER.

4591
Computer 46
kCAIPI: Reduction of interleaved 3D acquisition into a set of 2D simultaneous multi-slice (SMS) reconstruction problems
Gilad Liberman1 and Benedikt A Poser1

1Faculty of Psychology and Neuroscience, Maastricht University, Maastricht, Netherlands

The Fourier Transform (FT) of a vector of N=N1⋅N2 elements is decomposable into N1 FTs of N2-sized vectors followed by N2 FTs of N1-sized vectors, a fact utilized iteratively to produce the Fast FT algorithm. Put in MRI terminology, reconstructing N=k⋅M slices from k-undersampled kz-stacked trajectory can be achieved by FT, followed by solution of the M SMS problems of k slices. This can be used to reduce such 3D reconstruction problems into SMS problems, reducing memory and computational demands. The observation extends to CAIPI patterns. We term this approach kCAIPI.

4592
Computer 47
A fast approach for estimation of Spark of the sensing matrix for Compressed Sensing applications.
Bhairav Bipin Mehta1, Mingrui Yang2, and Mark Alan Griswold1

1Radiology, Case Western Reserve University, Cleveland, OH, United States, 2Department of Biomedical Engineering, Cleveland Clinic, Cleveland, OH, United States

Compressed sensing (CS) has been extensively used with wide spread application in MRI and other signal processing fields. Spark of the sensing matrix is at the heart of the CS framework for determining the success of the signal recovery for a given designed CS system. However, estimation of Spark of the sensing matrix is a combinatorial process, thus, practically difficult to estimate for realistic sizes of sensing matrices. The purpose of this work is to present a new optimization-problem-based approach for estimation of the Spark of the sensing matrix which will overcome the existing limitations, thereby, a tool to assess and design CS framework based systems. 

4593
Computer 48
Scalable self-calibrated interpolation of non-Cartesian data with GRAPPA
Seng-Wei Chieh1, Mostafa Kaveh1, Mehmet Akcakaya1,2, and Steen Moeller2

1Electrical and Computer Engineering, University of Minnesota-Twin Cities, Minneapolis, MN, United States, 2Center for Magnetic Resonance Research, Minneapolis, MN, United States

Conventional non-Cartesian parallel imaging reconstruction in k-space necessitates large amounts of calibration data for successful estimation of region-specific interpolation kernels. In this work, we propose a self-calibration strategy for obtaining region-specific non-Cartesian interpolation kernels from a single calibration dataset. This enables simple and efficient high-quality reconstruction of non-Cartesian parallel imaging.

4594
Computer 49
Regularized CG-SENSE for 30-channel 23Na head MRI at 7T
Melanie Schellenberg1, Armin M. Nagel1,2, Peter Bachert1, Mark E. Ladd1, and Nicolas G. R. Behl1

1Medical Physics in Radiology, German Cancer Research Center, Heidelberg, Germany, 2Institute of Radiology, University Hospital Erlangen, Erlangen, Germany

23Na MRI provides important information for many pathologies. However, its low SNR entails low spatial resolutions and long acquisition times. The proposed work reconstructs 3D radially undersampled in vivo 30-channel 23Na head data at B0=7T with a sensitivity encoding using a nonlinear conjugate gradient method (CG-SENSE) including a total variation and a discrete cosine transform. With CG-SENSE using iteratively generated Lagrangian coil sensitivities, image quality and contrast within the object are improved compared to sum of squares (SOS) and adaptive combination (ADC) reconstructions.


Artifacts, Implants & Corrections

Exhibition Hall
Thursday 9:15 - 10:15
 Acquisition, Reconstruction & Analysis

4595
Computer 51
Reconstruction of Undersampled Radial Free-breathing 3D Abdominal MRI using Conditional Generative Adversarial Network
Jun Lv1 and Jue Zhang2,3

1School of Computer and Control Engineering, Yantai University, Yantai, China, 2b. Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, China, 3c. College of Engineering, Peking University, Beijing, China

Free-breathing 3D abdominal imaging is challenging since respiratory motion can produce image blurring and ghosting artifact. Our purpose is to employ a novel deep learning method using conditional generative adversarial network (GAN) to reconstruct the undersampled radial 3D abdominal MRI. The whole network combines a generator G consists of 8 convolutional layers and corresponding 8 deconvolutional layers with a discriminator D which is formed using 11 convolutional layers. The GAN-based reconstructed images achieve similar quality to the ground-truth images. Additionally, the average reconstruction time is negligible. Therefore, this method can be adopted for a wide range of clinical applications.

4596
Computer 52
Artifact correction in spiral trajectory with high gradient performance
Daehun Kang1, Uten Yarach1, Joshua D Trzasko1, Myung-Ho In1, Erin M Gray1, Ek Tsoon Tan2, Nolan K Meyer1, Thomas K Foo2, Yunhong Shu1, John Huston1, and Matt A Bernstein1

1Department of Radiology, Mayo Clinic, Rochester, MN, United States, 2GE Global Research, Niskayuna, NY, United States

A low-cryogen, compact 3T MRI is equipped with high performance gradients, of which increased maximum gradient amplitude and slew rate can improve MR image quality of spiral trajectory to reduce susceptibility and off-resonance effect. However, use of the higher slew rate and gradient strength with an Archimedean spiral trajectory can lead to rotation artifacts and a local blurring. In this work, we corrected those artifacts with using a dynamic field camera and with attention to the azimuthal Nyquist sampling criterion.

4597
Computer 53
Wall-bounded Divergence-free Smoothing for Denoising of Velocity Data Measured by 4D Flow MRI
Chaehyuk Im1, Seungbin Ko1, Jeesoo Lee1,2, Jee-Hyun Cho3, Doosang Kim4, Sang Hyung Lee5, and Simon Song1,2

1Dept. of Mechanical Engineering, Hanyang University, Seoul, Korea, Republic of, 2Institute of Nano Science and Technology, Hanyang University, Seoul, Korea, Republic of, 3Bioimaging Research Team, Korea Basic Science Institute, Cheongju, Korea, Republic of, 4Thoracic and Cardio-vascular Surgery, Veterans Health Service Medical Center, Seoul, Korea, Republic of, 5Neurosurgery, SMG-SNU Boramae Medical Center, Seoul, Korea, Republic of

Flow data measured by 4D flow MRI often result in inaccurate wall shear stress estimation due to near-wall noise in velocity measurements. We propose wall-bounded divergence-free smoothing (WB-DFS) to denoise the flow data. This method minimizes a residual error under the divergence-free condition for a wall-bounded flow and simultaneously performs data smoothing. The denoising performance of WB-DFS was found to be the best among methods reported in literature. For example, the divergence of the velocity field was reduced by 95.6%. This method is expected to be helpful in accurately estimating wall shear stress in a vascular flow imaging. 

4598
Computer 54
Non-Cartesian MRI Systems Integrated Development using GPI and MATLAB – A New Rosette Pulse Sequence Example
Nanyque A Boyd1, Alexey V Dimov2, Amit R Patel2,3, Jacob P Goes1, Hui Wang4, Timothy J Carroll2, and Keigo Kawaji1,3

1Biomedical Engineering, Illinois Institute of Technology, Chicago, IL, United States, 2Radiology, The University of Chicago, Chicago, IL, United States, 3Medicine, The University of Chicago, Chicago, IL, United States, 4Philips, Gainesville, FL, United States

The default MRI scanner systems are not equipped with key technical resources for rapid deployment of novel non-Cartesian pulse sequence approaches. Here, we describe a Graphical Programming Interface-based (GPI) platform that is further embedded into the vendor reconstruction environment. This allows for comprehensive development and validation of new trajectories by integrating on-line MRI systems development with off-line resources such as MATLAB, which also enhances trainee-driven research efforts. This tool offers a set of resources including real-time display of MRI k-space and prototyping/characterization of sampling trajectory corrections that may simplify and streamline these non-Cartesian designs.

4599
Computer 55
Novel calibration-free correction of field inhomogeneity artifacts in EPI using a structured low rank method
Arvind Balachandrasekaran1, Merry Mani2, and Mathews Jacob1

1Electrical and Computer Engineering, University of Iowa, Iowa City, IA, United States, 2Radiology, University of Iowa, Iowa City, IA, United States

Echo Planar Imaging (EPI) is widely used in many dynamic imaging studies due to its capability to provide very good temporal resolution. However, the off-resonance artifacts due to long read out result in poor correspondence with structural scans and make data interpretation difficult. Here we introduce a novel framework, where the problem of artifact correction is transformed into a recovery of image time series from undersampled measurements. We exploit the exponential structure of the signal at every pixel along with the spatial smoothness of inhomogeneity map to recover the image series. Preliminary results demonstrate the potential of the proposed method.

4600
Computer 56
Elimination of fold-in artefacts for gradient inserts by using the existing whole-body gradient in synergy
Edwin Versteeg1, Abel Romeijnders1, Jeroen Hendrikse1, Dennis Klomp1, and Jeroen Siero1,2

1University Medical Centre Utrecht, Utrecht, Netherlands, 2Spinoza Center for Neuroimaging, Amsterdam, Netherlands

The short encoding field of a gradient insert makes such a coil susceptible to fold-in artefacts, especially when operated along the z-direction. We propose a method that almost completely eliminates this fold-in artefact by using the whole-body z-gradient as pre-winder and gradient insert (also in z) as readout gradient. This causes signal from outside the linear region of the gradient insert to stay dephased, thus suppressing the signal that folds in.  The proposed method is validated and quantified in simulation, and in experiments using a lightweight gradient insert that features a short encoding field.

4601
Computer 57
Streak Artifact Suppression in Radial MRI by Automatic Coil Selection
Lexiaozi Fan1,2, Hassan Haji-Valizadeh1,2, and Daniel Kim1,2

1Biomedical Engineering, Northwestern University, Evanston, IL, United States, 2Radiology, Northwestern University, Chicago, IL, United States

Streak artifact is very common in radial sampled images. One way we can reduce the artifact is to remove individual streaky coils by visually identification. Although it may not be a hard work, it’s time-consuming, especially when it comes to a large number of images. This abstract aims at developing an algorithm that can automatically detect these streaky coils, and suppress streak artifacts in reconstructed images.

4602
Computer 58
Metal Artifacts Reduction in DWI using Point Spread Function (PSF) Encoding
Sisi Li1, Yishi Wang1,2, Zhangxuan Hu1, Li Guan3, Yong Hai3, Xiao Han4, Yu Ma5, Hanwen Zhang3, Xiaolong Chen3, and Hua Guo1

1Department of Biomedical Engineering, Center for Biomedical Imaging Research, Beijing, China, 2Philips Healthcare, Beijing, China, Beijing, China, 3Chao Yang Hospital, Beijing, China, 4Beijing Jishuitan Hospital, Beijing, China, 5Tsinghua University Yu Quan Hospital, Beijing, China

  Though metal artifacts have been well-resolved in anatomical imaging by three dimensional multispectral imaging (3D-MSI) methods, diffusion weighted imaging (DWI) near metallic implants still remains a challenge, impeding various clinical applications. Point-Spread-Function encoded EPI (PSF-EPI) combined with Tilted-CAIPI can achieve highly accelerated distortion- and blurring-free high resolution DWI. By using an additional phase encoding, artifacts induced by severe susceptibility inhomogeneity around metal can be reduced even under a high acceleration rate. The reliable performance of PSF-EPI technique in metal artifacts reduction in DWI is demonstrated on phantom, in vitro swine forearm and in vivo patients.

4603
Computer 59
Comparison of Accelerated MAVRIC-SL with Robust-PCA and Conventional MAVRIC-SL in Evaluation of Symptomatic Total Hip Arthroplasties
Zoe Doyle1, Daehyun Yoon1, Philip Kenneth Lee1, Brian Hargreaves1, Christopher Beaulieu1, and Kathryn Stevens1

1Stanford University, Stanford, CA, United States

The substantial reduction of off-resonance artifacts near metal by multi-spectral imaging sequences facilitates the postoperative use of MRI to evaluate total hip arthroplasty patients, but its long scan time can be difficult for patients to tolerate. A novel MAVRIC-SL method using robust principal component analysis (RPCA) recently showed 2.6-fold reduced scan time with comparable artifact suppression. In this study we compare a conventional MAVRIC-SL method with the RPCA-accelerated MAVRIC-SL method in 36 total hip arthroplasty cases. Our data demonstrate nearly equivalent clinical sensitivity of the RPCA MAVRIC-SL method to the conventional method with a mild loss of spatial resolution.

4604
Computer 60
Pileup Artifact Correction Near Metal Implants Using Deep Neural Networks and Spectral K-Space Modulation
Kevin Koch1, Robin Karr1, and Andrew Nencka1

1Radiology, Medical College of Wisconsin, Milwaukee, WI, United States

Three-dimensional multispectral imaging (3D-MSI) techniques used for metal artifact correction can provide relatively clear images near most metallic implants.   However, within localized regions near some implants, 3D-MSI demonstrate residual artifacts that are unlike any other artifact previously seen in MR images. These confounding features in 3D-MSI are known as “pileup” or “ring” artifacts.  In this study, we present a novel approach to residual artifact correction in 3D-MSI that relies on 1) deep neural networks, 2) physical modeling of local gradients, and 3) k-space modulation and replacement of spectral data in compromised regions

4605
Computer 61
2D Imaging near Metallic Implants at 0.5T using High Time-Bandwidth Product RF pulses
Chad T Harris1, Andrew T Curtis1, Ian RO Connell2, Philip J Beatty1, Jeff A Stainsby1, and Curtis N Wiens1

1Research and Development, Synaptive Medical, Toronto, ON, Canada, 2Physics and Astronomy, Western University, London, ON, Canada

There are several benefits to reducing the main magnetic field strength including a reduction of imaging artefacts near metallic implants and the ability to significantly increase the peak B1+ of the RF pulses due to the reduction in SAR penalty. This enables higher time-bandwidth product (TBP) for a given RF pulse duration. In this work, we utilized high TBP RF pulses on a high-efficiency transmit coil and a 0.5T MR system to reduce through-plane distortions caused by metallic implants.  In addition to characterizing through-plane distortions, the impact of these pulses on in-plane distortions and SAR were also characterized.

4606
Computer 62
Clinical application of MAVRIC-SL in reducing metal implant artifacts in anterior cruciate ligament reconstruction
Jingyi Zhu1, Songbai Li1, Fei Bie2, and Lizhi Xie3

1Radiology Department of China Medical University First Hospital, Shenyang, China, 2MR Application China, GE Healthcare, Shen Yang, China, 3GE Healthcare China, Beijing, China

Some anterior cruciate ligament reconstructions have metal implants, and metal scrap may remain in the surgical procedure. Metals in the conventional magnetic resonance sequence, especially in the fat suppression sequences, produce large artifacts that affect the observation of the surrounding structure. This study performed conventional sequences and MAVRIC-SL sequence scan for patients with metal implants after ACLR and analysis the images. Conclusions that the oblique sagittal MAVRIC-SL PDWI FS sequence can be used to assisting in the diagnosis of traditional oblique sagittal T2WI FS and PDWI sequence.

4607
Computer 63
Novel use of the MAVRIC metal artifact reduction technique in MRI of the brain
Nathaniel Swinburne1, Regina Loccisano1, Duane Nicholson1, Norbert Ward1, Nayna Patel1, and Robert Young1

1Memorial Sloan Kettering Cancer Center, New York, NY, United States

While multiacquisition variable-resonance image combination (MAVRIC) is a recognized technique for metal artifact reduction in muskuloskeletal MRI, it is not widely described for MRI of the central nervous system.  We investigate the value of this technique for MRI of the brain in patients with MR conditional metal implants and find that MAVRIC-T1 is superior to conventional FSE T1 in both qualitative and quantitative metrics.

4608
Computer 64
The influence of B0 drift on the performance of the PLANET method and an algorithm for correction
Yulia Shcherbakova1, Cornelis A.T. van den Berg2, Chrit T.W. Moonen1, and Lambertus W. Bartels1

1Center for Image Sciences, Imaging Division, UMC Utrecht, Utrecht, Netherlands, 2Department of Radiotherapy, Imaging Devision, UMC Utrecht, Utrecht, Netherlands

The PLANET method has been recently proposed to quantify the relaxation parameters T1 and T2, the banding free magnitude, the local off-resonance ∆f0, and the RF phase from RF phase-cycled balanced steady-state free precession (bSSFP) data. The PLANET model requires a static B0 field over the course of the acquisition. However, due to gradient activity, B0 drift can happen. In this work we present a study of the influence of B0 drift on the performance of the method and we propose a strategy for correction.

4609
Computer 65
Gaussianization of Diffusion MRI Magnitude Data Using Spatially Adaptive Phase Correction
Feihong Liu1,2, Geng Chen2, Jun Feng1, Pew-Thian Yap2, and Dinggang Shen2

1Northwest University, Xi'an, China, 2University of North Carolina at Chapel Hill, Chapel Hill, NC, United States

We present a method of effective phase-correction of diffusion-weighted images with the goal of obtaining real-valued signals with zero-mean Gaussian distributed noise. Our method estimates the noise level locally and is hence well-suited for spatially-varying noise.

4610
Computer 66
Eliminating susceptibility induced hyperintensities in ultra highd field T1w MPRAGE brain images
Ruoyun Emily Ma1, Thomas Henry2, and Pierre-François Van de Moortele1

1Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States, 2Department of Neurology, University of Minnesota, Minneapolis, MN, United States

Ultra high field brain MPRAGE images are commonly affected by local susceptibility induced hyperintensities, which are pronounced in inferior frontal lobe and inferior temporal lobe. In this work, we propose a straightforward approach by applying a frequency offset of 300Hz and widening the bandwidth by 40% to the hyperbolic secant inversion pulse provided by the standard MPRAGE sequence, to eliminate this artefact without introducing additional incomplete inversion through the brain. This approach was tested across different subjects and proven to have robust performance in artefact elimination against variable local frequency offsets.

4611
Computer 67
B0 shim improvement in the inferior frontal lobe by head-tilting: feasibility and comparison with 3rd order shimming
Seulki Yoo1,2, Hayoung Song1,2, Won Mok Shim1,2, and Seung-Kyun Lee1,2

1Department of Biomedical Engineering, Sungkyunkwan University, Suwon, Korea, Republic of, 2Center for Neuroscience Imaging Research, IBS, Suwon, Korea, Republic of

Susceptibility-induced signal dropout and image quality impairment in the gradient-echo based imaging are well known problems in brain MRI at high fields. Here, we experimentally demonstrate the feasibility and benefit of head-tilted brain scan as a means to reduce B0 inhomogeneity and associated gradient echo signal loss in the inferior frontal lobe (IFL), and compare the shim improvement with simulated 3rd order shimming in the whole brain.

4612
Computer 68
Water-fat separation in spiral readout acquisition for the liver using the convolutional neural network: an approach to reduce blurring artifacts
Daiki Tamada1, Ryoichi Kose2, Katsumi Kose2, Hiroshi Onishi1, and Utaroh Motosugi1

1Department of Radiology, University of Yamanashi, Chuo, Japan, 2MRIsimulations Inc., Tokyo, Japan

Water/fat separation algorithm for two-echo spiral acquisition in the liver was developed using the convolutional neural network (CNN). The processing in the CNN was performed in the sinogram domain. A Bloch simulator was used to simulate the phase error in the k-space caused by the off-resonance components of background and fat. A volunteer study showed the successful water/fat separation using the proposed method without additional echoes of reference scans.

4613
Computer 69
Simultaneous multislice EPI reconstruction by incorporating split slice-GRAPPA with slice-dependent 2D Nyquist ghost correction (PEC-SP-SG)
Yilong Liu1,2, Markus Barth3, Mengye Lyu1,2, and Ed X. Wu1,2

1Laboratory of Biomedical Imaging and Signal Processing, The University of Hong Kong, Hong Kong, China, 2Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China, 3Center for Advanced Imaging, The University of Queensland, Brisbane, Australia

Simultaneous multislice (SMS) EPI reconstruction is challenging due to slice-dependent 2D phase differences between opposite polarities, which is collapsed across slices. Additionally, slice leakage is one major concern in some applications including diffusion and functional MRI. The proposed SMS EPI reconstruction incorporates phase error correction with split slice-GRAPPA (PEC-SP-SG), and was evaluated using simulation, phantom and in vivo experiments. Results show that the proposed approach can offer a robust SMS EPI reconstruction with slice-dependent 2D Nyquist ghost correction, and provide a balance between slice leakage and in-plane artifacts.

4614
Computer 70
Detection and Correction of MR EPI Data Corrupted by Spike Noise
Ziyi Pan1, He An2, Elaine Yuen Phin Lee2, and Hua Guo1

1Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China, 2Department of Diagnostic Radiology, Li Ka Shing Fadiology, Li Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong, China

EPI-based MR images are prone to the spike artifact, which is usually caused by small electrical discharges i.e. sparks that emit radio frequency power within the bandwidth of the scanner receiving system 1 during the MR acquisition. Spike noise causes ripples or stripes covered on the object and can hamper the qualitative or quantitative analysis of the MR images. In this work, we developed a reliable technique that combines Robust Principal Component Analysis 2 (RCPA) with median filtering to robustly detect and correct spike-affected images in human pelvic diffusion weighted imaging (DWI). The overall image quality and the lesion conspicuity were improved after spike removal.

4615
Computer 71
Cascaded Deep Learning Networks for Automated Image Quality Evaluation of Structural Brain MRI
SHEEBA SUJIT1, REFAAT GABR1, IVAN CORONADO1, and PONNADA NARAYANA1

1Department of Diagnostic and Interventional Imaging, University of Texas Health Science Center, Houston, TX, United States

Visual quality assessment of MRI is subjective and impractical for large datasets. In this study, we present a cascaded convolutional neural network (CNN) model for automated image quality evaluation of structural brain MRI. The multisite Autism Brain Imaging Data Exchange dataset of ~1000 subjects was used to train and evaluate the proposed model. The model performance was compared with expert evaluation. The first network rated individual slices, and the second network combined the slice ratings into a final image score. The network achieved 74% accuracy, 69% sensitivity, and 74% specificity, demonstrating that deep learning can provide robust image quality evaluation.

4616
Computer 72
Automatic Quality Assessment of Pediatric MRI via Nonlocal Residual Neural Networks
Siyuan Liu1, Kim-Han Thung1, Weili Lin1, Pew-Thian Yap1, Dinggang Shen1, and UNC/UMN Baby Connectome Project Consortium2

1Biomedical Research Imaging Center, University of North Carolina at Chapel Hill, Chapel Hill, NC, United States, 2Univerisity of North Carolina at Chapel Hill, Chapel Hill, NC, United States

Manual MRI quality assessment is time-consuming, subjective, and error-prone. We show that image quality of contrast-varying pediatric MR images can be automatically assessed using deep learning with near-human accuracy.

4617
Computer 73
MR Optimum – A web-based application for signal-to-noise ratio evaluation.
Eros Montin1,2, Roy Wiggins1,2, Kai Tobias Block1,2, and Riccardo Lattanzi1,2,3

1Department of Radiology, Bernard and Irene Schwartz Center for Biomedical Imaging, New York University School of Medicine, New York, NY, United States, 2Department of Radiology, Center for Advanced Imaging Innovation and Research (CAI2R), New York University School of Medicine, New York, NY, United States, 3Sackler Institute of Graduate Biomedical Sciences, New York University School of Medicine, New York, NY, United States

The signal-to-noise ratio (SNR) is a commonly used metric to evaluate image quality and radiofrequency coil performance in MRI. However, its calculation could be challenging. Here we introduce MR Optimum, a novel web-based application for the evaluation of SNR. By means of a user-friendly web GUI, readily available via any internet browser, it provides access to various methods for SNR calculation. The computing unit can be installed on a local server or distributed over the cloud. Results can be visualized, analyzed and exported in various formats. MR Optimum could help standardizing how SNR is calculated and reported in scientific publications. 

4618
Computer 74
An Investigation into the Origins of an MRI Artifact Induced by Increasing Temperature
Ryan T. Oglesby1,2, Wilfred W. Lam2, and Greg J. Stanisz1,2

1Medical Biophysics, University of Toronto, Toronto, ON, Canada, 2Physical Sciences, Sunnybrook Health Sciences Centre, Toronto, ON, Canada

An artifact has been observed in pure water samples after increasing the temperature above 25 °C, regardless of the heating mechanism. This study aims to determine the origins of the temperature-induced artifact by using MR thermometry and T1 relaxation to investigate samples containing increasing concentrations of agar. The addition of a small concentration (0.1%) of agar eliminates the temperature artifact suggesting that the increased viscosity of the samples decreases convection currents. Moreover, this addition of a small concentration of agar study provides a practical means of experimentally scanning samples at physiological temperature.

4619
Computer 75
General Abnormality Detection in MR Images using a Generative Adversarial Network
Karsten Sommer1, Michael Helle1, Axel Saalbach1, Marvin Rühe1,2, and Heinrich Schulz1

1Philips Research, Hamburg, Germany, 2University of Applied Sciences Karlsruhe, Karlsruhe, Germany

In this study, a Generative Adversarial Network is used for detection of abnormalities in MR brain images such as lesions, artifacts etc. Given a query image, a generative model that is trained to create normal appearing brain images is used to find a best match. Since abnormalities cannot be reproduced accurately by the generative model, pathologies and artifacts become apparent.


New RF & Gradient Strategies

Exhibition Hall
Thursday 9:15 - 10:15
 Acquisition, Reconstruction & Analysis

4620
Computer 76
On the Effective Centre of Excitation and the Point of Gradient Moment Expansion for 2D-Selective Excitation in the Presence of Flow
Clarissa Wink1, Simon Schmidt2, Jean Pierre Bassenge1,3, Sandy Szermer1, Giulio Ferrazzi1, Bernd Ittermann1, Tobias Schaeffter1, and Sebastian Schmitter1

1Physikalisch-Technische Bundesanstalt (PTB), Braunschweig and Berlin, Germany, 2Medical Physics in Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany, 3Working Group on Cardiovascular Magnetic Resonance, Experimental and Clinical Research Center, a joint cooperation between the Charité Medical Faculty and the Max-Delbrueck Center for Molecular Medicine, Berlin, Germany

In this work, we demonstrate the distinction and importance of two virtual time points during excitation for correct flow compensation and quantification: the centre of excitation ($$$t_0^\text{m}$$$) at which spins are excited and thus magnitude is generated, and the isophase time-point ($$$t_0^\text{ph}$$$) at which all excited spins are in phase. A general method to determine $$$t_0^\text{m}$$$ is presented and $$$t_0^\text{ph}$$$ and $$$t_0^\text{m}$$$ are shown to be not necessarily identical. Finally, phantom experiments demonstrate that the knowledge of $$$t_0^\text{m}$$$ is required to remove the displacement artefact in phase-encoding directions to enable correct flow compensation and imaging.

4621
Computer 77
Quantum-Inspired RF Pulse Optimization
Sherry Huang1, Darryl C. Jacob2,3, Michael Beverland3, Stephen Jordan3, Helmut G. Katzgraber3, Matthias Troyer3, Rasim Boyacioglu4, Yun Jiang4, Dan Ma4, Mark A. Griswold4, Julie Love3, and Debra F. McGivney4

1Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 2Physics and Astronomy, Texas A&M University, College Station, TX, United States, 3Microsoft, Redmond, WA, United States, 4Radiology, Case Western Reserve University, Cleveland, OH, United States

RF pulse design is crucial in creating the desired magnetization profile which is the basis of Magnetic Resonance Imaging. There are various methods to generate the RF pulse and gradient waveforms based on Fourier relationships, filter design, or optimizations. These methods rely on assumptions and approximations due to computational power constraints. Here we present preliminary results of using quantum inspired algorithms for Bloch simulation and RF pulse design optimization.

4622
Computer 78
Improved Accuracy of FLASH-based B1+ Mapping by Optimization of the Fourier Encoding Matrix
Omer F. Oran1, L. Martyn Klassen1, and Ravi S. Menon1,2

1Centre for Functional and Metabolic Mapping, University of Western Ontario, London, ON, Canada, 2Department of Medical Biophysics, University of Western Ontario, London, ON, Canada

FLASH-B1 mapping with a Fourier-encoding scheme works only for a limited range of flip-angles due to pronounced saturation effects that occur for short TR. The transmit voltage can be adjusted to satisfy this requirement in problematic Fourier combinations that have high dynamic range. However, this results in an unnecessary reduction of the dynamic range in combinations that are already within the accepted range. This study addresses this problem by optimizing the Fourier encoding matrix such that the flip-angle is reduced only in regions of high flip-angle in the problematic combinations.

4623
Computer 79
BEEEP: B1-robust Energy Efficient Excitation Pulses
Eric Van Reeth1, Hélène Ratiney1, Olivier Beuf1, Soukaina Kanice1, Steffen J Glaser2, and Dominique Sugny3

1CREATIS, Villerubanne, France, 2Department of Chemistry, Technical University of Munich, Garching, Germany, 3Laboratoire Interdisciplinaire Carnot de Bourgogne, Dijon, France

This study introduces a new family of broadband B1-robust excitation (90°) pulses for MRI with large enough bandwidth (+/- 1 kHz) to account for static field inhomogeneities, and minimal energy deposition. RF pulses are designed with a regularized optimal control algorithm, which is able to adapt the pulse B1-robustness range to fit the coil limits in terms of peak amplitude and energy. In vitro acquisitions using an endoluminal-shaped RF transmit coil show comparable excitation profiles than BIR4 pulses, although BEEEP pulses deposit 5.2 times less energy.

4624
Computer 80
A Simple Method for Constrained Optimal Control RF Pulse Design
Sydney N. Williams1, Jon-Fredrik Nielsen2, Jeffrey A. Fessler3, and Douglas C. Noll2

1Imaging Centre of Excellence, University of Glasgow, Glasgow, United Kingdom, 2Biomedical Engineering, University of Michigan, Ann Arbor, MI, United States, 3Electrical Engineering and Computer Science, University of Michigan, Ann Arbor, MI, United States

Optimal control (OC) methods for RF pulse design are useful in cases where the small-tip angle (STA) approximation is violated. Furthermore, designs with physically meaningful constraints (e.g., RF peak amplitude and integrated power) eliminate the need for parameter tuning to create realizable pulses. In this abstract we introduce a constrained fast OC method that easily generalizes to a variety of RF pulse designs. We demonstrate with examples of SMS and spectral prewinding pulses in simulation and in vivo. The constrained fast OC method guarantees that RF pulses will meet physical constraints while outperforming their non-OC counterparts.

4625
Computer 81
RF pulse design via time optimal control for combined excitation, refocusing and inversion
Christoph Stefan Aigner1, Armin Rund2, Christina Graf1, Karl Kunisch2, and Rudolf Stollberger1

1Institute of Medical Engineering, Graz University of Technology, Graz, Austria, 2Institute for Mathematics and Scientific Computing, University of Graz, Graz, Austria

This work demonstrates a constrained joint design of minimum duration RF pulse and slice selective gradient waveforms for combined SMS excitation, refocusing and inversion scenarios. A hybrid trust-region semismooth Newton/quasi-Newton method with exact derivatives via adjoint calculus is used to solve the time optimal problem on fine spatial and temporal grids. Specific hardware and safety constraints, including maximal RF, slice selective gradient, slew rate amplitudes as well as global SAR estimates, guarantee practical applicability. High-resolution GRE, crushed SE and inversion recovery GRE slice profile measurements on a 3T MR system validate the numerical results.

4626
Computer 82
PASTeUR: Package of Anatomical Sequences using parallel Transmission UniveRsal kT-point pulses
Vincent Gras1, Franck Mauconduit2, Alexandre Vignaud1, Caroline Le Ster1, Lisa Leroi1, Alexis Amadon1, Eberhard Pracht3, Markus Boland3, Rüdiger Stirnberg3, Tony Stöcker3, Benedikt A. Poser4, Christopher Wiggins5, Xiaoping Wu6, Kamil Ugurbil6, and Nicolas Boulant1

1Neurospin, CEA, Université Paris-Saclay, Gif-sur-Yvette, France, 2Siemens Healthineers, Saint Denis, France, 3German Center for Neurodegenerative Diseases (DZNE), Bonn, Germany, 4Department of Cognitive Neuroscience, Faculty of Psychology and Neuroscience, Maastricht University, Maastricht, Netherlands, 5Scannexus, Maastricht, Netherlands, 6Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States

Despite its power to counteract the inevitable radiofrequency field inhomogeneity problem at ultra-high field, parallel transmission has failed to be embraced by the community in routine due to a cumbersome workflow. Universal pulses have shown great potential to circumvent this problem by providing plug and play solutions. Here we validate a package of 3D anatomical sequences for a given commercial coil covering multiple contrasts for use in clinical routine and including, thanks to their versatility, very few pulse solutions. The utilization of universal kT-points  enables direct embedding of these pulses in the sequences and easy handling of the power/SAR limits.

4627
Computer 83
3D Selective RF and Gradient Waveforms designed by using a GPU Accelerated Genetic Algorithm
Christopher Mirfin1, Paul Glover1, and Richard Bowtell1

1Sir Peter Mansfield Imaging Centre, School of Physics & Astronomy, University of Nottingham, Nottingham, United Kingdom

The Genetic Algorithm (GA) is motivated by the process of natural selection, allowing mutliple initialisations. Due to the stochastic nature of genetic algorithms they are beneficial in avoiding local minima, although they can require significantly more function evaluations to run than a traditional solver. In this work, motivated by the field of shape optimisation, an approach is taken to perform the joint design of RF and gradient waveforms using a GA with a GPU-accelerated iterative solver.

4628
Computer 84
Spin Lock Adiabatic Correction (SLAC) of BIR4 pulses for increased B1-insensitivity at 7T
Edward M Green1,2, Yasmin Blunck1,2, James C Korte3, Bahman Tahayori4,5, Peter M Farrell6, and Leigh A Johnston1,2

1Dept. of Biomedical Engineering, University of Melbourne, Melbourne, Australia, 2Melbourne Brain Centre Imaging Unit, University of Melbourne, Melbourne, Australia, 3Department of Physical Sciences, Peter MacCallum Cancer Centre, Melbourne, Australia, 4Department of Medical Physics and Biomedical Engineering, Shiraz University of Medical Sciences, Shiraz, Iran (Islamic Republic of), 5Dept. of Medical Physics and Biomedical Engineering, Shiraz University of Medical Sciences, Shiraz, Iran (Islamic Republic of), 6Dept. of Electrical and Electronic Engineering, University of Melbourne, Melbourne, Australia

Inhomogeneous B1 excitation impedes image quality, particularly at high field.  Adiabatic pulse modulation ameliorates this effect, however super-adiabatic properties can be exploited to further improve performance.  Spin Lock Adiabatic Correction (SLAC) pulses can be applied to any adiabatic pulse shape, through reduction of flip angle inaccuracies induced by B1 variability.  In this work, SLAC is derived for BIR4 pulse shapes, and the superior performance of SLAC-BIR4 is demonstrated in both simulation and phantom experiments at 7T.   The SLAC procedure is an attractive analytical alternative to numerical optimisation of adiabatic pulses.

4629
Computer 85
Sensing low-frequency, low-amplitude AC magnetic fields at ultra-low field with steady-state SIRS
Bragi Sveinsson1,2,3, Neha Koonjoo1,2,3, Thomas Witzel1,2, and Matthew Rosen1,2,3

1Martinos Center, Massachusetts General Hospital, Charlestown, MA, United States, 2Harvard Medical School, Boston, MA, United States, 3Physics, Harvard University, Cambridge, MA, United States

In this work, we demonstrate a method to detect low-frequency, low-amplitude AC magnetic fields in an ultra-low-field (ULF) MRI system using a steady-state implementation of the Stimulus-Induced Rotary Saturation (SIRS) method. The method optimizes SNR efficiency by applying the SIRS mechanism in a bSSFP scan. This approach takes advantage of the low SAR and small absolute B0 deviations of the ULF system. We describe simulation results, show a clear signal response in phantoms, and describe an in vivo protocol for using the method to detect response from an auditory stimulus.

4630
Computer 86
PAREWISE:  High-bandwidth, reduced-rephase, slice-selective excitation pulses for high-field spectroscopy
James B Murdoch1

1Canon Medical Research USA, Mayfield Village, OH, United States

To meet the localization needs of high-field spectroscopy, new high-bandwidth computer-optimized RF excitation pulses have been generated, featuring phase modulation, t=0 points near the end of the pulse duration, B1 insensitivity over a ±15% range, and sharp excitation profiles.  In the design process, a maximum amplitude of 1 kHz was imposed, and bandwidth was increased by stepping up the pulse length from 5 to 11 ms.  Both 90° and 60° flip angles have been investigated thus far.

4631
Computer 87
3D Inner Volume Imaging with 3D Tailored Outer Volume Suppression RF pulses
Tianrui Luo1, Douglas Noll1, and Jon-Fredrik Nielsen1

1fMRI Lab, Univ. of Michigan., Ann Arbor, MI, United States

3D inner-volume (IV) steady-state imaging is a candidate for, e.g., high-resolution BOLD fMRI, but it can be challenging to achieve sufficient outer-volume (OV) signal suppression. This is particularly true for 3D tailored RF pulses that excite an arbitrary 3D IV (e.g., a cylinder of finite height) thus enabling fast non-Cartesian readouts, as 3D IV pulses are more difficult to design than 2D tailored pulses. We propose to insert a 3D tailored OV suppression pulse to help suppress OV steady-state signal in 3D IV imaging sequences that use 3D tailored IV excitation pulses. We show that this strategy can substantially improve the IV signal profile for commonly used and emerging steady-state sequences such as spoiled gradient-echo (SPGR), balanced SSFP, and small-tip fast-recovery (STFR).

4632
Computer 88
Spiral RARE with annular segmentation
Juergen Hennig1, Marius Menza1, Antonia Barghoorn1, Bruno Riemenschneider1, Stefan Kroboth1, and Maxim Zaitsev1

1Dept. of Radiology, Medical Physics, University Medical Center Freiburg, Freiburg, Germany

We present a new approach to spiral RARE with annular segmentation. Annular segmentation leads to monotonous T2-dependent weighting of signal amplitudes across k-space and thus to very benign artifact behavior. Preliminary results show that single shot images (128x128) of decent quality can be acquired without fat suppression and without field inhomogeneity correction.  By cyclic shifting of the spiral segments quantitative T1- and/or T2- images can be acquired in a few seconds. Sequence implementation was performed swiftly and efficiently in MatLab with the vendor independent Pulseq sequence development environment.

4633
Computer 89
A Novel Approach to Investigate 1D TRASE MRI Pulse Sequence Performance in Imperfect B1 Fields
Pallavi Bohidar1, Hongwei Sun2, Gordon Sarty1, and Jonathan Sharp2

1University of Saskatchewan, Saskatoon, SK, Canada, 2University of Alberta, Edmonton, AB, Canada

The Transmit Array Spatial Encoding (TRASE) MRI technique uses transmit radio-frequency (RF) magnetic field (B1) phase gradients for spatial encoding. Imaging performance is reliant on |B1| homogeneity. This study investigates the performance of a set of variants of 1-dimensional TRASE under conditions of |B1| inhomogeneity. Both Bloch equation simulations and experimental confirmations are presented. Results show that one variant of the sequence allows imaging over ~90% of the FOV for Δ|B1| = ± 15% allowing the use of practical B1 transmit coil designs. This improves upon the previous sequence which generated ~75% usable field-of-view (FOV).

4634
Computer 90
AZTEK: Adaptive Zero TE K-space trajectories
Tanguy Boucneau1,2,3, Brice Fernandez4, Anne Menini5, Florian Wiesinger6, Luc Darrasse1,2,3, and Xavier Maître1,2,3

1IR4M, CNRS, Orsay, France, 2IR4M, Univ. Paris-Sud, Orsay, France, 3IR4M, Université Paris-Saclay, Orsay, France, 4Applications & Workflow, GE Healthcare, Buc, France, 5Applications & Workflow, GE Healthcare, Menlo Park, CA, United States, 6Applications & Workflow, GE Healthcare, Garching bei München, Germany

Because of short signal lifetimes and respiratory motion, 3D MR lung imaging is still challenging today. Zero TE (ZTE) pulse sequences are promising as they overcome the problem of short T2*. Nevertheless, because of the continuous readout gradients they require, their k-space trajectories are non-optimal for retrospective gating. We propose AZTEK, a 3D radial trajectory featuring several tuning parameters to adapt the acquisition to any moving organ while keeping a smooth transition between consecutive spokes. The increase in image quality was validated with static and moving phantom experiments, and demonstrated with dynamic thoracic imaging performed on a human volunteer.

4635
Computer 91
Redesigned Variable-Density Cones Trajectory for High Resolution MR Imaging
Kwang Eun Jang1,2, Srivathsan P Koundinyan1, Dwight G Nishimura1, and Shreyas S Vasanawala3

1Magnetic Resonance Systems Research Lab (MRSRL), Department of Electrical Engineering, Stanford University, Stanford, CA, United States, 2Department of Bioengineering, Stanford University, Stanford, CA, United States, 3Department of Radiology, Stanford University, Stanford, CA, United States

The 3D cones trajectory has been employed for various applications. In this work we transform the task of designing the cones trajectory into a generic procedure of discretizing an analytic coordinate. We present a new discretization scheme with a spiral path on a unit sphere, which enables allocation of readout interleaves on distinct conic surfaces for any given number of readouts. Subsequently, we derive the relationship between the sampling density of each interleaf and that of overall interleaves, which allows us matching the sampling density of the cones trajectory to the 1/f-model of 3D images.

4636
Computer 92
Silent Structural Imaging and T1-mapping with a Rapid-Radial Twice-Prepared (R2P2) Sequence
Emil Ljungberg1, Florian Wiesinger1,2, and Tobias C Wood1

1Neuroimaging, King's College London, London, United Kingdom, 2ASL Europe, GE Healthcare, Munich, Germany

We combined the MP2RAGE sequence with a silent radial ZTE readout and acquired a high-contrast, high-SNR T1-weighted image and quantitative T1 map at 0.9mm isotropic resolution at 3T free from B1-inhomogeneity. This has potential for high resolution structural imaging of populations that would not otherwise tolerate MRI due to the acoustic noise of standard sequences.


4637
Computer 93
Improving pseudo-continuous arterial spin labelling at ultra-high field using a VERSE-guided parallel transmission strategy
Yan Tong1, Peter Jezzard1, Thomas W Okell1, and William T Clarke1

1Wellcome Centre for Integrative Neuroimaging, FMRIB Division, Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, United Kingdom

Implementing ASL at ultra-high field is challenging due to increased specific absorption rate (SAR), along with B1+ and B0 inhomogeneity. Parallel transmission (pTx) provides additional degrees of freedom to mitigate B1+ inhomogeneity. Among various pTx strategies, RF shimming is a simple formulation that modulates the complex weights of each RF channel. In this study, we explored the possibility of using VERSE to further improve PCASL at 7T, and VERSE-guided RF shimming was shown to achieve improved SNR in perfusion-weighted images over power-matched Circularly-Polarised (CP) mode and RF-shimmed Gaussian labelling schemes.

4638
Computer 94
Quiet Dixon Imaging with Looping Star Sequence
Holger Eggers1, Kay Nehrke1, Peter Börnert1, and Johan van den Brink2

1Philips Research, Hamburg, Germany, 2Philips Healthcare, Best, Netherlands

Dixon imaging with a conventional bipolar multi-gradient-echo sequence is often loud, mainly because of rapid switching of the strong readout gradient. In this work, the feasibility of using the Looping Star sequence instead is explored, which was recently introduced for quiet radial multi-gradient-echo imaging. Different variants of a dual-acquisition Looping Star sequence are proposed and demonstrated to allow a robust water-fat separation in phantom and volunteer experiments.


4639
Computer 95
Universal Parallel Transmit Pulse Design for 3-Dimensional Local-Excitation – A 9.4T Simulation Study
Ole Geldschläger1, Tingting Shao1, and Anke Henning1

1Hochfeld-Magnetresonanz, Max-Planck-Institut für biologische Kybernetik, Tübingen, Germany

This study introduces a parallel-transmission (pTx) radio-frequency (RF) pulse-design-method to create an universal pTx RF-pulse that excites the same 3-dimensional local excitation pattern with a desired flip-angle in different human heads at 9.4T. Thus, it prospectively abandons the need for time-consuming subject specific B1+ mapping and pTx-pulse calculation during the scan session. The resulting universal pulses created magnetization profiles with an only marginal worse Normalized-Root-Mean-Square-Error (NRMSE) compared to the magnetization profiles produced by the pulses tailored to individual heads.

4640
Computer 96
A K-means Clustering Algorithm for MRI Virtual Observation Points Compression in Local SAR Supervision
Xianglun Mao1, Joseph V. Rispoli1,2, and David J. Love1

1Electrical and Computer Engineering, Purdue University, West Lafayette, IN, United States, 2Weldon School of Biomedical Engineering, Purdue University, West Lafayette, IN, United States

Virtual observation point (VOP) compression has become a standard technique to address SAR and temperature constraints in MRI parallel transmission (pTx) design. SAR matrices need to be pre-averaged for the regions of interest, and finally be conservatively compressed to a much smaller set of SAR matrices (i.e. VOPs) that is still capable of reliably calculating a peak spatial SAR. We demonstrated a new approach that used a k-means algorithm for VOP compression. The new VOP compression method does not yield any under-estimation but allows for a lower over-estimation in the peak local SAR prediction.  

4641
Computer 97
Aliased Coil Compression
Gilad Liberman1 and Benedikt A Poser1

1Faculty of Psychology and Neuroscience, Maastricht University, Maastricht, Netherlands

Cartesian sub-sampling patterns play a major role in routine MRI, usually reconstructed using GRAPPA or SENSE and image based regularization. Coil compression is commonly applied to reduce computational load and noise. Software coil compression achieves only mediocre compression factors without compromising signal. Geometrical/ESPIRiT coil-compression use fully-sampled axes, when availables, to improve compression factors without reducing signal or reconstruction level. In this work we present Aliased Coil Compression for Cartesian subsampling patterns, achieving optimal compression without any signal loss. The method is especially useful for alleviating fast image-domain regularization (compressed sensing or deep learning) for available sequences.

4642
Computer 98
Zoomed 3D GRE EPI utilizing a Segmented 2D Selective RF Pulse Excitation
Naoharu Kobayashi1, Michael Mullen1,2, Kamil Ugurbil1, and Michael Garwood1

1Center for Magnetic Resonance Research, Department of Radiology, University of Minnesota, Minneapolis, MN, United States, 2School of Physics and Astronomy, University of Minnesota, Minneapolis, MN, United States

Segmented 2D selective RF pulse excitation is introduced in 3D-EPI for zoomed imaging. The feasibility of the 2D selective pulse segmentation was tested in phantom and in vivo brain measurements. The segmented pulse excitation provided nearly identical excitation profile to a non-segmented pulse excitation. In zoomed in vivo brain imaging, the segmented pulse excitation showed conspicuous improvement of susceptibility artifacts around the frontal sinus.

4643
Computer 99
Rapid multi-dimensional RF pulse design with deep learning
Mads Sloth Vinding1, Birk Skyum2, Ryan Sangill1, and Torben Ellegaard Lund1

1Department of Clinical Medicine, Center of Functionally Integrative Neuroscience, Aarhus University, Aarhus, Denmark, 2Interdisciplinary Nanoscience Center, Aarhus University, Aarhus, Denmark

For multi-dimensional RF pulses, neural networks and deep learning may boost the clinical applicability by allowing very rapid pulse predictions, based on offline training and offline generated training libraries. This can potentially offer opportunities, for example, to revive slow, abandoned pulse design techniques, or to include many more constraints or complexities into the pulse designs that until now were infeasible to bring into a clinical setting, since the neural network will simply learn the features of the training library. We are demonstrating the principle with numerical simulations, and phantom and in vivo experiments.


Machine Learning for Image Reconstruction: A New Frontier

Exhibition Hall
Thursday 9:15 - 10:15
 Acquisition, Reconstruction & Analysis

4644
Computer 101
An Unsupervised Deep Learning Approach for Reconstructing Arterial Spin Labeling Images from Noisy Data
Kuang Gong1, Paul Kyu Han1, Debra E. Horng1, Georges El Fakhri1, Chao Ma1, and Quanzheng Li1

1Gordon Center for Medical Imaging, Massachusetts General Hospital and Harvard Medical School, Boston, MA, United States

Recently convolutional neural networks (CNNs) have been successfully applied to computer vision tasks and attracted growing interests in medical imaging. One barrier for the application of deep neural networks is the need of large amounts of training pairs, which are not always available in clinical practice. Inspired by the deep image prior method, this work presents a new image reconstruction framework based on CNN representation where no training pairs and pre-training are needed. We demonstrate the effectiveness of the proposed method by performing denoising and image reconstruction using noisy arterial spin labeling (ASL) data with and without undersampling.

4645
Computer 102
Unsupervised Learning for Improved Fidelity Multi-contrast MRI
Ke Wang1, Frank Ong1, Jonathan I Tamir1, and Michael Lustig1

1Electrical Engineering and Computer Sciences, University of California, Berkeley, Berkeley, CA, United States

Multi-contrast MRI acquisitions from a single scan have tremendous potential to streamline exams and reduce imaging time. However, maintaining clinically feasible scan times necessitates significant undersampling, pushing the limits on compressed sensing and other low-dimensional techniques. While learning methods have been proposed to overcome this limitation, they rely on fully sampled data for training, which are difficult to obtain for multi-dimensional imaging. Here, we present an unsupervised learning approach based on convolutional sparse coding, which learns a structured convolutional dictionary directly from undersampled k-space datasets.  We apply the proposed method to T2 Shuffling knee datasets and demonstrate improvements to image sharpness and relaxation dynamics compared to the locally low-rank reconstruction.

4646
Computer 103
Signal Stability and Sensitivity of Referenceless Reconstructions by a Neural Network in Simultaneous Multi-Slice Imaging
Klaus Eickel1,2, Martin Blaimer3, and Matthias Günther1,2

1MR-Imaging & Spectroscopy, Faculty 01 (Physics/Electrical Engineering), University of Bremen, Bremen, Germany, 2MR Physics, Fraunhofer MEVIS, Bremen, Germany, 3Fraunhofer IIS, Würzburg, Germany

A deep neural network for reconstruction of SMS data without the need of additional reference data to calibrate for the spatial encoding information of the multi-coil receiver is presented. Noise-propagation through the reconstruction process is investigated in form of g-factors. Simulations with pseudo-multiple replicas showed robustness and stability of this new method. In addition, the sensitivity for physiological signal variations of this approach is considered in BOLD-signal dynamics. Results are compared to conventional methods like split slice-GRAPPA.

4647
Computer 104
LORAKI: Reconstruction of Undersampled k-space Data using Scan-Specific Autocalibrated Recurrent Neural Networks
Tae Hyung Kim1, Pratyush Garg1,2, and Justin P. Haldar1

1Electrical Engineering, University of Southern California, Los Angeles, CA, United States, 2Electrical Engineering, Indian Institute of Technology (IIT) Kanpur, Kanpur, India

We introduce LORAKI, a novel MRI reconstruction method that bridges two powerful existing approaches (LORAKS and RAKI).  Like RAKI (a deep learning extension of GRAPPA), LORAKI trains a scan-specific autocalibrated convolutional neural network (which only relies on autocalibration data, and does not require external training data) to interpolate missing k-space samples.  However, unlike RAKI, LORAKI is based on a recurrent convolutional neural network architecture that is motivated by the iterated convolutional structure of a certain LORAKS algorithm.  LORAKI is very flexible and can accommodate arbitrary k-space sampling patterns.  Experimental results suggest LORAKI can have better reconstruction performance than state-of-the-art methods.

4648
Computer 105
Unpaired Super-Resolution GANs for MR Image Reconstruction
Ke Lei1, Morteza Mardani1,2, Shreyas Vasawanala2, and John Pauly1

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Radiology, Stanford University, Stanford, CA, United States

While undersampled MRI data is easy to obtain, lack of high-quality labels for dynamic organs impedes the common supervised training of deep neural nets for MRI reconstruction. We propose an unpaired training super-resolution model with pure GAN loss to use a minimal amount of labels but all available low-quality data for training. Leveraging Wasserstein-GANs with gradient penalty followed by a data-consistency refinement high-quality Knee MR images are recovered from 3-fold undersampled single coil measurements using 20% of the labels compared with a paired training model. 

4649
Computer 106
Semi-Supervised Learning for Reconstructing Under-Sampled MR Scans
Feiyu Chen1, Joseph Y Cheng2, John M Pauly1, and Shreyas S Vasanawala2

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Radiology, Stanford University, Stanford, CA, United States

Supervised deep-learning approaches have been applied to MRI reconstruction, and these approaches were demonstrated to significantly improve the speed of reconstruction by parallelizing the computation and using a pre-trained neural network model. However, for many applications, ground-truth images are difficult or impossible to acquire. In this study, we propose a semi-supervised deep-learning method, which enables us to train a deep neural network for MR reconstruction without using fully-sampled images.

4650
Computer 107
Deep learning aided compressed sensing for accelerated cardiac cine MRI
Tobias Wech1, Thorsten A. Bley1, and Herbert Köstler1

1Department of Diagnostic and Interventional Radiology, University Hospital Würzburg, Würzburg, Germany

A reconstruction technique for accelerated functional cardiac MRI is presented that exploits a convolutional neural network trained for semantic segmentation of undersampled data. The idea is inspired by the experience that the human eye is capable of distinguishing between typical undersampling artifacts and cardiac shape and/or motion, even for high acceleration factors. The temporal courses of the segmentations determined by the network are used for an efficient sparsification within a compressed sensing algorithm.

4651
Computer 108
W-net: A Hybrid Compressed Sensing MR Reconstruction Model
Roberto Souza1,2 and Richard Frayne1,2

1Radiology and Clinical Neuroscience, Hotchkiss Brain Institute, University of Calgary, Calgary, AB, Canada, 2Seaman Family MR Centre, Foothills Medical Centre, Calgary, AB, Canada

Compressed sensing (CS) magnetic resonance (MR) imaging acquisitions reduce MR exam times by decreasing the amount of data acquired during acquisition, while still reconstructing high quality images. Deep learning methods have the advantage of reconstructing images in a single step as opposed to iterative (and slower) CS methods. Our proposal aims to leverage information from both k-space and image domains, in contrast to most other deep-learning methods that only use image domain information. We compare our W-net model against four recently published deep-learning-based methods. We achieved second best results in the quantitative analysis, but with more visually pleasing reconstructions. 

4652
Computer 109
Deep Scaled Domain Learning for Compressed MRI using Optional Scaling Transform
Satoshi ITO1 and Kohei SATO1

1Utsunomiya University, Utsunomiya, Japan

Image domain learning designed for image denoiser has superior performance when aliasing artifacts are incoherent; however, its performances will be degraded if the artifacts show small incoherency. In this work, a novel image domain learning CNN is proposed in which images are transformed to scaled space to improve the incoherency of artifacts. Simulation and experiments showed that the quality of obtained image was fairly improved especially for lower sampling rate and the quality was further improved by cascaded network. It was also shown that the resultant PSNR exceeded one of the transform learning method.

4653
Computer 110
FlowNet: High-Speed Compressed Sensing 4D Flow MRI Image Reconstruction using Loop Unrolling
Jonas Walheim1, Valery Vishnevskiy 1, and Sebastian Kozerke1

1Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland

A variational neural network for the reconstruction of compressed sensing 4D flow MRI is presented.  Nine iterations of an iterative reconstruction are unfolded in a neural network which was trained using eight retrospectively undersampled datasets. A phase-invariant network architecture was designed with two types of filter operations, one with equal real and imaginary component and the other operating on image magnitude only. The method is shown to outperform spatial regularization in the Wavelet domain. A retrospectively undersampled patient scan demonstrates that the network can reconstruct pathologies based on healthy training samples. Reconstruction of prospectively undersampled 4D flow MRI shows good agreement of peak velocities and peak flow.

4654
Computer 111
Multi-scale Unrolled Deep Learning Network for Accelerated MRI
Ukash Nakarmi1,2, Joseph Yitan Cheng1,2, Edgar Anselmo Rios Piedra1,2, Morteza Mardani1,2, John M Pauly2, Leslie Ying3,4, and Shreyas Vasanawala1

1Department of Radiology, Stanford University, Stanford, CA, United States, 2Department of Electrical Engineering, Stanford University, Stanford, CA, United States, 3Department of Electrical Engineering, University at Buffalo, Buffalo, NY, United States, 4Department of Biomedical Engineering, University at Buffalo, Buffalo, NY, United States

Model prior based reconstruction and data-centric prior reconstruction are two strong paradigms in image reconstruction inverse problems. In this abstract, we propose a framework that integrates the model prior and data-centric multi-scale deep learning priors for reconstructing magnetic resonance images (MRI) from undersampled k-space data. The proposed framework brings together the best of both paradigms and has proven superior to conventional accelerated MRI reconstruction techniques. 

4655
Computer 112
Dual-domain Generative Adversarial Model for Accelerated MRI Reconstruction
Guanhua Wang1,2, Enhao Gong2,3, Suchandrima Banerjee4, Karen Ying5, Greg Zaharchuk6, and John Pauly2

1Biomedical Engineering, Tsinghua University, Beijing, China, 2Electrical Engineering, Stanford University, Stanford, CA, United States, 3Subtle Medical, Menlo Park, CA, United States, 4Global Applied Science Laboratory, GE Healthcare, Menlo Park, CA, United States, 5Engineering Physics, Tsinghua University, Beijing, China, 6Radiology, Stanford University, Stanford, CA, United States

Previous CS frameworks based on Deep Learning like GANCS have demonstrated improved quality and efficiency. To further improve the restoration of the high-frequency details and the suppression of aliasing artifacts, a data-driven regularization is explicitly added on the k-space, in the form of an adversarial loss (GAN). In this work, the cross-domain generative adversarial model is trained and evaluated on diverse datasets and show decent generalization ability. For both quantitative comparison and visual inspection, the proposed method achieves better reconstruction than previous networks.

4656
Computer 113
Dynamic Multi-Coil Reconstruction using Variational Networks
Kerstin Hammernik1, Matthias Schloegl2, Erich Kobler1, Rudolf Stollberger2,3, and Thomas Pock1

1Institute of Computer Graphics and Vision, Graz University of Technology, Graz, Austria, 2Institute of Medical Engineering, Graz University of Technology, Graz, Austria, 3BioTechMed-Graz, Graz, Austria

In this work, we present a variational network for reconstructing dynamic multi-coil data. Incorporation of parallel imaging increases the acceleration potential due to additional spatial information, but was not considered so far in other learning-based reconstruction approaches for dynamic MRI. We show that variational network reconstructions with learned spatio-temporal regularization lead to further improvements in image quality compared to state-of-the-art Compressed Sensing approaches for different CINE cardiac datasets and acceleration factors with 10-times faster reconstruction time.

4657
Computer 114
Pseudo-Cartesian k-Space Interpolation Using Artificial Neural Networks
Nikolai J Mickevicius1, Eric S Paulson1, and Andrew S Nencka2

1Radiation Oncology, Medical College of Wisconsin, Milwaukee, WI, United States, 2Radiology, Medical College of Wisconsin, Milwaukee, WI, United States

This work aims to extend the RAKI method for artificial intelligence-based k-space interpolation to non-Cartesian acquisitions. It was tested in radial acquisitions up to acceleration factors of 7. This method performs similarly well, or better than total-variation regularized sensitivity encoding.

4658
Computer 115
A Reference-Free Convolutional Neural Network Model for Magnetic Resonance Imaging Reconstruction from Under-Sampled k-Space
Yang Song1, Yida Wang1, Xu Yan2, Minxiong Zhou3, Bingwen Hu1, and Guang Yang1

1Shanghai Key Laboratory of Magnetic Resonance, East China Normal University, Shanghai, China, 2MR Scientific Marketing, Siemens Healthcare, Shanghai, China, 3Shanghai University of Medicine & Health Sciences, Shanghai, China

We used a reference-free model based on convolutional neural network (RF-CNN) to reconstruct the under-sampled magnetic resonance images. The model was trained without fully sampled image (FS) as the reference. We compared our model with the traditional compressed sensing reconstruction (CS) and the CNN model trained by FS. Mean square error and structure similarity were used to evaluate the model. Our RF-CNN model performed better than CS, but did not perform as good as usual CNN model. 

4659
Computer 116
Parallel Imaging in Time-of-Flight Magnetic Resonance Angiography Using Deep Multi-Stream Convolutional Neural Networks
Yohan Jun1, Taejoon Eo1, Hyungseob Shin1, Taeseong Kim1, Hojoon Lee2,3, and Dosik Hwang1

1Electrical Electronic Engineering, Yonsei University, Seoul, Korea, Republic of, 2Radiology and Research Institute of Radiological Science, Yonsei University College of Medicine, Seoul, Korea, Republic of, 3Department of Radiology, Inje University College of Medicine, Busan, Korea, Republic of

A deep parallel imaging network (“DPI-net”) was developed to reconstruct 3D multi-channel MRA from undersampled data. It comprises two deep-learning networks: a network of multi-stream CNNs for extracting feature maps of multi-channel images and a network of reconstruction CNNs for reconstructing images from the multi-stream network output feature maps. DPI-net was effective in reconstructing 3D time-of-flight MRA from highly undersampled multi-channel MR data, achieving superior performance, both quantitatively and qualitatively, over conventional parallel imaging and other deep-learning methods.

4660
Computer 117
Parallel Imaging based on k-x Domain Interpolation using Deep Neural Networks
Hyungseob Shin1, Taejoon Eo1, Yohan Jun1, Taeseong Kim1, and Dosik Hwang1

1Electrical and Electronic Engineering, Yonsei University, Seoul, Korea, Republic of

 In this study, we compare two deep learning approaches to reconstruct multi-channel magnetic resonance (MR) images subsampled along phase-encoding (PE) direction. They are both based on the Fully-Connected (FC) layers but are performed in two different domains : Image domain, and k-x domain which is 1D inverse Fourier transformed (IFT) k-space. We demonstrate that the latter method shows superior performance to the former one in terms of removing the aliasing artifacts and recovering the details of MR images. The performance of the proposed method to the conventional MR reconstruction on image domain was qualitatively and quantitatively evaluated.

4661
Computer 118
Non-Cartesian k-space Deep Learning for Accelerated MRI
Yoseob Han1 and Jong Chul Ye1

1Korea Advanced Institute of Science and Technology (KAIST), Daejeon, Korea, Republic of

The annihilating filter-based low-rank Hankel matrix approach (ALOHA) [1] is one of the most recent compressed sensing (CS) approaches that directly interpolates the missing k-space data using low-rank Hankel matrix completion. Inspired by the recent low-rank Hankel matrix decomposition using data-driven framelet basis [2], we propose a completely data-driven deep learning algorithm for k-space interpolation. In particular, our method can be applied directly by simply adding an additional re-gridding layer to non-Cartesian k-space trajectories such as radial trajectories.

4662
Computer 119
Accelerated 3D Non-Cartesian Reconstruction with Deep Learning
Mario O. Malavé1, Srivathsan P. Koundinyan1, Christopher M. Sandino1, Frank Ong2, Joseph Y. Cheng1,3, and Dwight G. Nishimura1

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Electrical Engineering and Computer Sciences, University of California, Berkeley, CA, United States, 3Radiology, Stanford University, Stanford, CA, United States

In this work, we demonstrate the application of a non-Cartesian unrolled architecture in reconstructing images from undersampled 3D cones datasets. One shown application of this method is for reconstructing undersampled 3D image-based navigators (iNAVs), which enable monitoring of beat-to-beat nonrigid heart motion during a cardiac scan. The proposed non-Cartesian unrolled network architecture provides similar outcomes as l1-ESPiRIT in one-twentieth of the time, and the reconstructions exhibit robustness when using an undersampled 3D cones trajectory.  

4663
Computer 120
Data Consistency Networks for (Calibration-less) Accelerated Parallel MR Image Reconstruction
Jo Schlemper1, Jinming Duan2, Cheng Ouyang1, Chen Qin1, Jose Caballero1, Joseph V. Hajnal3, and Daniel Rueckert1

1Department of Computing, Imperial College London, London, United Kingdom, 2Faculty of Medicine, Institute of Clinical Sciences, Imperial College London, London, United Kingdom, 3Biomedical Engineering, King's College London, London, United Kingdom

We present simple reconstruction networks for multi-coil data by extending deep cascade of CNN’s and exploiting the data consistency layer. In particular, we propose two variants, where one is inspired by POCSENSE and the other is calibration-less. We show that the proposed approaches are competitive relative to the state of the art both quantitatively and qualitatively.

4664
Computer 121
Fast estimation of GRAPPA kernel using Meta-learning
Dongwook Lee1 and Jong Chul Ye1

1Bio and Brain Eng., KAIST, Daejeon, Korea, Republic of

This paper proposes an accelerated MR reconstruction method for parallel imaging from uniformly undersampled k-space data by learning scan-specific GRAPPA kernel using the long short-term memory network (LSTM). In particular, the meta-leaner LSTM is redesigned to quickly estimate the GRAPPA kernel for each k-space from its auto-calibration signals (ACS). The proposed method shows improved reconstruction performance with minimum error.

4665
Computer 122
Deep Plug-and-Play Prior for Parallel MRI Reconstruction
Ali Pour Yazdanpanah1, Onur Afacan1, and Simon K. Warfield1

1Computational Radiology Laboratory, Boston Children's Hospital and Harvard Medical School, Boston, MA, United States

Fast data acquisition in Magnetic Resonance Imaging (MRI) is vastly in demand and scan time directly depends on the number of acquired k-space samples. Conventional MRI reconstruction methods for fast MRI acquisition mostly relied on different regularizers which represent analytical models of sparsity. However, recent data-driven methods based on deep learning has resulted in promising improvements in image reconstruction algorithms. In this paper, we propose a deep plug-and-play prior framework for parallel MRI reconstruction problems which utilize a deep neural network (DNN) as an advanced denoiser within an iterative method. We demonstrate that a deep plug-and-play prior framework for parallel MRI reconstruction with a regularization that adapts to the data itself results in excellent reconstruction accuracy and outperforms the clinical gold standard GRAPPA method.

4666
Computer 123
A Cascaded Residual Dense Network for Cardiac MR Imaging
Ziwen Ke1,2, Shanshan Wang2, Cheng Li2, Huitao Cheng1,2, Leslie Ying3, Xin Liu2, Hairong Zheng2, and Dong Liang1,2

1Research center for Medical AI, Shenzhen Institutes of Advanced Technology, Shenzhen, China, 2Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Shenzhen, China, 3Department of Biomedical Engineering and Department of Electrical Engineering, The State University of New York, Buffalo, NY, United States

Cardiac MR imaging plays an important role in clinical diagnosis. But the long scan time limits its wide applications. To accelerate data acquisition, deep learning based methods have been applied to effectively reconstruct the undersampled images. However, current deep convolutional neural network (CNN) based methods do not make full use of the hierarchical features from different convolutional layers, which impedes their performances. In this work, we propose a cascaded residual dense network (C-RDN) for dynamic MR image reconstruction with both local features and global features being fully explored. Our proposed C-RDN achieves the best performance on in vivo datasets compared to the iterative optimization methods and the state-of-the-art CNN method.

4667
Computer 124
Reconstruction of high undersampling rate images using a cascade of convolutional neural networks
Xi Chen1, Shuo Chen2, and Rui Li2

1Beijing Institute of Technology, Beijing, China, 2Center for Biomedical Imaging Research, Department of Biomedical Engineering, School of Medicine, Tsinghua University, Beijing, China

Imaging speed is important in many magnetic resonance imaging (MRI) applications because long scan time increases the risk of artifacts. At present, reconstruction method based on compressed sensing and deep learning significantly increases the speed of MRI scan. However, the performance of current models is not good at high undersampling rate. Here we used a large dataset to improve the undersampling rate of a CNN based MR reconstruction while maintaining high image quality. Our results showed an average 2.6% root-mean-square error in reconstructing from 16-fold undersampling k-space, which outperforms traditional method.

4668
Computer 125
Deep MRI Reconstruction without Ground Truth for Training
Peizhou Huang1, Chaoyi Zhang2, Hongyu Li2, Sunil Kumar Gaire2, Ruiying Liu2, Xiaoliang Zhang1, Xiaojuan Li3, and Leslie Ying1,2

1Biomedical Engineering, State University of New York at Buffalo, Buffalo, NY, United States, 2Electrical Engineering, State University of New York at Buffalo, Buffalo, NY, United States, 3Program of Advanced Musculoskeletal Imaging (PAMI), Cleveland Clinic, Cleveland, OH, United States

Deep learning has recently been applied to image reconstruction from undersampled k-space data with success. Most existing works require both undersampled data and ground truth image as the training pair. It is not practical to obtain a large number of ground truth images for training in some MR applications. Here a novel deep learning network is studied for image reconstruction using only undersampled data for training. Experiment results demonstrate the feasibility of training without the ground truth images for image reconstruction.


Improving Definition & Reducing Artifacts

Exhibition Hall
Thursday 9:15 - 10:15
 Acquisition, Reconstruction & Analysis

4669
Computer 126
High-Resolution Simultaneous Mapping of Brain Function and Metabolism
Rong Guo1,2, Yibo Zhao1,2, Yudu Li1,2, Yao Li3, and Zhi-Pei Liang1,2

1Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States, 2Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL, United States, 3Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China

We present a new method for simultaneous mapping of brain function and metabolism. This method provides an unprecedented capability to simultaneously obtain high-resolution metabolic maps (2.4×2.4×3.0 mm3) and brain functional maps (3.0×3.0×2.6 mm3) of the whole brain coverage (230×230×120 mm3) in 8 minutes. The proposed method extends the subspace-based imaging framework of the SPICE technique with a new data acquisition scheme and exploits the complementary information between MRSI and fMRI signals for high-quality image reconstruction. Brain imaging experiments have been carried out, demonstrating the impressive capability of our method. With further improvement, the method can provide an unprecedented tool for mapping brain function and metabolism simultaneously. 

4670
Computer 127
CSF-suppressed T2 weighted imaging at 7T
Jullie W Pan1, Chan Moon1, Arun Antony2, Frank Lieberman2, Vikas Agarwal1, and Hoby P Hetherington1

1University of Pittsburgh, Pittsburgh, PA, United States, 2UPMC, Pittsburgh, PA, United States

T2-weighted lesional imaging is most commonly performed using inversion recovery turbo spin echoes. At 7T, however, this acquisition is limited for specific absorption rate and resolution. We implement a strategy that uses a driven equilibrium spin-echo preparation within an inversion recovery with multiple 3D gradient-echo imaging blocks to generate CSF-suppressed T2 weighted sensitivity. Images are combined using the self-normalization approach. Data acquired with an 8x2 transceiver array are shown to demonstrate sensitivity in brain tumors and epilepsy.

4671
Computer 128
Visualizing the Myocardium in-vivo with a 3D uTE Acquisition
Larry Kasuboski1, Jason Ortman2, Sho Tanaka3, Bharath Ambale Venkatesh2, and Joao A Lima2

1Canon Medical Research USA, Mayfield Village, OH, United States, 2Johns Hopkins University School of Medicine, Baltimore, MD, United States, 3Canon Medical Systems Corporation, Otawara-shi, Japan

A dark blood 3D uTE acquisition scheme with an MSDE pre-pulse is shown to provide suppression of flowing blood, while maintaining good definition of the myocardium.

4672
Computer 129
Optimize Wave-CAIPI MPRAGE protocol for the study of Short-term Apparent Change (SAC) of Grey Matter in Motor Training
Tie-Qiang Li1, Tobias Granberg1, and Sven Petersson1

1Department of Medical Radiation and Nuclear Medicine, Karolinsak University Hospital, Huddinge, Sweden

To assess the extent and dynamics of short-term apparent change (SAC) of GM in motor training we investigated how the VBM results are affected by the different levels of acceleration of the MP-RAGE sequence using the wave-CAIPI technique which provides highly accelerated MPRAGE imaging and retain high image quality. The optimized wave-CAIPI MPRAGE imaging protocol overcomes the g-factor noise amplification penalty and allows for over an order of magnitude acceleration of MPRAGE imaging in VBM studies. The standard and wave-CAIPI MPRAGE sequences have different sensitivity in detecting SAC of GM likely due to their differences in noise and contrast characteristics.

4673
Computer 130
Simultaneous B1- and Fat-Corrected T1 Mapping Using Chemical-Shift Encoded MRI
Nathan Tibbitts Roberts1,2, Diego Hernando1,3, Timothy J Colgan1, Xiaoke Wang1,4, and Scott B Reeder1,3,4,5,6

1Radiology, University of Wisconsin - Madison, Madison, WI, United States, 2Electrical and Computer Engineering, University of Wisconsin - Madison, Madison, WI, United States, 3Medical Physics, University of Wisconsin - Madison, Madison, WI, United States, 4Biomedical Engineering, University of Wisconsin - Madison, Madison, WI, United States, 5Medicine, University of Wisconsin - Madison, Madison, WI, United States, 6Emergency Medicine, University of Wisconsin - Madison, Madison, WI, United States

Spatially varying B1 inhomogeneities and tissue fat are known to be confounders of quantitative T1 mapping methods that use multiple flip-angle techniques. Separately acquired B1 calibration maps can be used to correct flip angle errors caused by B1 inhomogeneities, but this requires an additional acquisition. In this work we propose a comprehensive approach that combines concepts from actual flip-angle imaging with variable flip-angle imaging to simultaneously estimate B1 inhomogeneity, T1, proton density fat-fraction and R2*. The feasibility and noise performance of this joint acquisition and fitting approach are evaluated using Cramer-Rao Lower Bound analysis, simulations, phantom experiments, and preliminary in vivo examples.

4674
Computer 131
Robust R1rho asymmetry performed with optimal B1 selection
Baiyan Jiang1, Jian Hou1, and Weitian Chen1

1Department of Imaging and Interventional Radiology, The Chinese University of Hong Kong, Hong Kong, Hong Kong

The measurement of R1rho (1/T1rho) spectrum and its asymmetry have several advantages over Chemical Exchange Saturation Transfer (CEST) and Chemical Exchange Spin-lock (CESL). It is recently becoming one of the important approaches for probing the chemical exchange process. In this work, we demonstrated the advantage of R1rho asymmetry over CESL, and proposed a way to predict the optimal B1 for R1rho asymmetry.

4675
Computer 132
Super-resolution MRI with 2D Phaseless Encoding
Rui Tian1, Franciszek Hennel1, and Klaas P Pruessmann1

1Institute for Biomedical Engineering, ETH Zurich and University of Zurich, Zurich, Switzerland

Super-resolution MRI with 1D phaseless encoding achieves high-resolution with immunity to shot-dependent phase fluctuation by simultaneously acquiring multiple k-space bands. We now explore a 2D extension of this technique to facilitate more k-space sampling strategies. Two distinct encoding schemes were analyzed and tested with EPI acquisition. By properly adjusting the overlapping of the mixed k-space bands, the 2D phaseless encoding could also be combined with the spiral acquisition. The amplitude modulation caused by band overlapping was eliminated by an inverse filter during reconstruction. The overlapped bands regions were also exploited to provide information about unexpected bands errors for post-processing corrections.

4676
Computer 133
Super-resolution based on the Signal Extrapolation in Phase scrambling Fourier Transform Imaging using Deep Convolutional Neural Network
Satoshi ITO1

1Utsunomiya University, Utsunomiya, Japan

The signal obtained in phase scrambling Fourier transform imaging can be extrapolated beyond sampling length after data acquisition like Half-phase encoding method. To realize the method for phase varied images, precise phase distribution map is required. In this paper, a new post-processing super resolution in PSFT imaging is proposed in which deep convolution neural network (CNN) is used and phase map is not required. Simulation and experimental results showed that spatial resolution was fairly improved with signal extrapolation and the improvement of spatial resolution is proportional to the strength of phase scrambling coefficient.

4677
Computer 134
Semi-Classical Signal Analysis Method with Soft-Thresholding for MRS denoising
Abderrazak Chahid1, Sourav Bhaduri2, Malik Wali1, Eric Achten3, Hacene Serrai 2,4, and Taous-Meriem Laleg-Kirati1

1Computer, Electrical and Mathematical Science and Engineering (CEMSE) division, King Abdullah University of Sciences and Technology (KAUST), Thuwal, Saudi Arabia, 2Department of Radiology and Nuclear Medicine, University of Ghent, Gent, Belgium, 3Department of Radiology, Department of Radiology and Nuclear Medicine, University of Ghent, Gent, Belgium, 4Robarts Research Institute, University of Western Ontario, London, ON, ON, Canada

A Semi-Classical Signal Analysis (SCSA)   method with soft thresholding is proposed for MRSI denoising. The SCSA takes advantage of the pulse-shaped MRS spectrum to decompose both real and imaginary parts, into localized basis given by squared eigenfunctions of the Schrödinger operator. An optimization-based soft-threshold is provided to find optimal semi-classical parameters, for both the real and imaginary parts of the MRS signal. The optimal SCSA parameters discard the eigenfunctions representing noise from the noisy spectrum, and conserve the eigenfunctions representing the useful information. The obtained in-vivo results show the efficiency of the SCSA with soft thresholding in removing noise and conserving metabolite signals.

4678
Computer 135
Encoding trajectory optimization based on the pixel variance using graphics processing units
Ali Sadr1, Stefan Kroboth1, Elmar Fischer1, Feng Jia1, Sebastian Littin1, Huijun Yu1, Jürgen Hennig1, and Maxim Zaitsev1

1Dept. of Radiology, Medical Physics, Medical Center University of Freiburg, Faculty of Medicine, University of Freiburg, Freiburg im Breisgau, Germany

Undersampled trajectories were optimized for parallel imaging with explicit consideration of the RF coil sensitivities in order to complement the RF coil elements. A second-order approximation of pixel variance was used as a metric to evaluate encoding trajectories and also serves as the cost function in the optimization problem, solved using Simulated Annealing. The metric was implemented on a Graphical Processing Unit (GPU) to accelerate computations. The developed method was evaluated on two test cases with Cartesian and radial sampling for isotropic and anisotropic fields-of-view. Resulting optimized trajectories led to improved image quality, more uniform SNR and reduced g-factors.

4679
Computer 136
Online compressed sensing MR image reconstruction for high resolution $$$T_2^*$$$ imaging
Loubna El Gueddari1,2, Emilie Chouzenoux3,4, Jean-Christophe Pesquet3, Alexandre Vignaud1, and Philippe Ciuciu1,2

1CEA/NeuroSpin, Gif-sur-Yvette, France, 2INRIA-CEA Parietal team, Univ. Paris-Saclay, Gif-sur-Yvette, France, 3CVN, Centrale-Supélec, Univ. Paris-Saclay, Gif-sur-Yvette, France, 4LIGM, Paris-Est University, Marne-La-Vallée, France

Compressed sensing theory reduces lengthy acquisition time in MRI at the expense of computationally demanding iterative reconstruction. Usually, reconstruction is performed offline once all the data have been collected. Here, we introduce an online CS reconstruction framework that interleaves acquisition and reconstruction steps in a convex setting and permits the delivery of intermediate images on the scanner console during acquisition. In particular, the sum of acquisition and reconstruction times is reduced without compromising image quality. The gain of this strategy is shown both on retrospective Cartesian and prospective non-Cartesian under-sampled ex-vivo baboon brain data at 7T with an in-plane resolution of 400$$$\mu$$$m.

4680
Computer 137
Radial Single-Shot Fast Spin Echo: Toward Fast, Motion-Robust, Multi-Contrast Imaging
Daniel V Litwiller1, Kang Wang2, Lloyd Estkowski3, Ali Ersoz4, Ty Cashen2, and Ersin Bayram5

1Global MR Applications and Workflow, GE Healthcare, New York, NY, United States, 2Global MR Applications and Workflow, GE Healthcare, Madison, WI, United States, 3Global MR Applications and Workflow, GE Healthcare, Menlo Park, CA, United States, 4GE Healthcare, Waukesha, WI, United States, 5Global MR Applications and Workflow, GE Healthcare, Houston, TX, United States

We present a radial Single-Shot Fast Spin Echo pulse sequence that is capable of generating multiple contrasts from a single spin echo train via temporal filtering.  Undersampling artifacts are minimized by using a long echo train and an aggressive variable refocusing flip angle scheme.  In vivo feasibility is demonstrated in the brain.

4681
Computer 138
Rapid Prototyping of Spiral Based Three-Points Dixon Acquisition and Reconstruction Using Pulseq
Marina Manso Jimeno1, Sairam Geethanath2, and John Thomas Vaughan2

1University of Groningen, Groningen, Netherlands, 2Columbia University in the city of New York, New York City, NY, United States

Spiral imaging possesses special characteristics that would benefit both research and clinical MR fields. However, its use is often limited due to the difficulties its implementation on a scanner entails. This problem frequently limits the spread and evolution of spirals and consequently their potential and multiple applications. The current study proposes a solution to this problem by demonstrating a rapid prototyping of a three-points Dixon spiral sequence. Results agree with the standard sequence used by the system manufacturer for the same purpose. The software used is open-source and makes possible the sequence implementation in multiple vendor scanners.

4682
Computer 139
Improvement of 3D-STIR for TRANCE non-contrast MR angiography at 3T using stretched adiabatic inversion pulse
Kosuke Morita1, Masami Yoneyama2, Masanobu Nakamura2, Takeshi Nakaura3, Seitaro Oda3, Masahiro Hatemura1, and Yasuyuki Yamashita3

1Radiology, Kumamoto University, Kumamoto-shi, Japan, 2Philips Japan, Tokyo, Japan, 3Diagnostic Radiology, Kumamoto University, Kumamoto-shi, Japan

The purpose of our study was to improve 3D-STIR for TRANCE non-contrast MR angiography in clinical 3.0T MR system using modified hyperbolic secant (HS4) pulses. the higher field strength poses additional challenges to 3D STIR, including wider offset frequency between water and fat combined with larger B0 and B1 inhomogeneities, which reduce the reliability of fat suppression. 3D STIR (TRANCE) with HS4 pulse has clearly improved fat suppression due to B0/B1 inhomogeneous compared with conventional HS pulse in clinical.

4683
Computer 140
3D Cartesian Fast-interrupted Steady-state Sequence (FISS) with Intrinsic Fat Suppression
Thomas Kuestner1, Aurelien Bustin1, Olivier Jaubert1, Radhouene Neji1,2, Claudia Prieto1, and René M Botnar1

1School of Biomedical Engineering & Imaging Sciences, King's College London, London, United Kingdom, 2Siemens Healthineers, Erlangen, Germany

Fat-suppressed balanced steady-state free precession (bSSFP) is a rapid imaging sequence often used in cardiovascular MRI. Recently a fast interrupted steady-state (FISS) sequence was proposed which periodically interrupts the steady-state of the bSSFP. The resulting frequency response modulation can be leveraged for suppression of the off-resonant fat signal without the need of fat preparation pulses. FISS was demonstrated for 2D radial acquisitions, however challenges to apply this approach to 3D Cartesian have been reported. Here we propose to extend FISS to 3D Cartesian imaging and investigate its behavior by extended phase graph simulations and in-vivo leg, abdominal measurements.

4684
Computer 141
Simultaneous Functional MRI and Proton Echo-Planar Spectroscopic Imaging in Human Brain (fPEPSI)
Stefan Posse1, Bruno Sa De La Rocque Guimaraes 1, Kishore Vakamudi2, and Steen Moeller3

1Neurology,Physics and Astronomy, University of New Mexico, Albuquerque, NM, United States, 2Neurology, University of New Mexico, Albuquerque, NM, United States, 3Center for Magnetic Resonance Research, Radiology, University of Minnesota, Minneapolis, MN, United States

This study introduces simultaneous fMRI and MRSI integrates multi-slab echo-volumar-imaging (MEVI) into the water suppression module of Proton-Echo-Planar-Spectroscopic-Imaging to simultaneously acquire fMRI, water suppressed and water reference data in a single scan (fPEPSI). FMRI image quality and BOLD sensitivity acquired with 4x4x6 mm3 resolution was comparable to our recently developed MEVI method. Spectral quality and sensitivity of 3D metabolite maps acquired with 4x4x7 mm3 resolution were comparable to conventional PEPSI. This hybrid fMRI/MRSI approach considerably reduces scan times in multi-modal clinical research studies and it is applicable to characterizing neurotransmitter and lactate concentration changes in relation to BOLD signal changes.

4685
Computer 142
Fixed-Point MR Imaging for Early Gliobrastoma Detection
Zhao Li1, Chao-Hsiung Hsu2, Huiyuan Zheng2, and Yung-Ya Lin2

1Chemistry and Biochemistry, UCLA, Los Angeles, CA, United States, 2UCLA, Los Angeles, CA, United States

Problem Early detection of high-grade malignancy, such as GBM, remains challenging using MRI.

Methods A new approach using continuous-wave and feedback field to reach “fixed-point spin dynamics” was developed to enhances the local magnetic-field gradient variations due to irregular water contents and deoxyhemoglobin concentration in early GBM.

Results In vivo MR images and mappings acquired on orthotopic GBM mice using “fixed-point pulse sequence” shows 3-4 times of enhancement in GBM contrast than the best conventional images acquired.

Conclusion Simulations and in vivo GBM mouse models validated the superior contrast/sensitivity and robustness of fixed-point spin dynamics method towards early GBM detection.


4686
Computer 143
Implications of within-scan patient head motion on B1+ homogeneity and Specific Absorption Rate at 7T
Emre Kopanoglu1, Alix Jean Deeley Plumley1, M. Arcan Erturk2, Cem M. Deniz3, and Richard G. Wise1

1CUBRIC, School of Psychology, Cardiff University, Cardiff, United Kingdom, 2Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States, 3Department of Radiology, School of Medicine, New York University, New York, NY, United States

Parallel-transmit pulses are commonly used to improve B1+-homogeneity at higher field strengths, while local-SAR constraints are applied to ensure safety. However, patient motion may become unavoidable with longer scans or less cooperative patients, and motion may affect B1+-homogeneity and local-SAR. We investigated the effect of all 6 degrees-of-freedom of head motion on B1+-homogeneity and local-SAR for parallel-transmit multi-spoke pulses using simulations. We observed more than a 2-fold increase in local-SAR due to motion for some pulses. We also investigated the changes in B1+-homogeneity of spokes pulses using in-vivo B1+-maps and showed regional variations between 12%-22% in the excitation profile.

4687
Computer 144
Symmetric Priors for Regularisation of Elastic Deformations (SPRED) - efficient GPU-accelerated enforcement of diffeomorphism in B-spline parametrised 3D nonlinear registration
Frederik J. Lange1, Stephen M. Smith1, and Jesper L. R. Andersson1

1Wellcome Centre for Integrative Neuroimaging, FMRIB Division, Nuffield Dept of Clinical Neurosciences, University of Oxford, Oxford, United Kingdom

We present a method for regularising nonlinear registration which produces deformations which are biologically more plausible than conventional techniques. Our method, the Symmetric Prior for Regularisation of Elastic Deformations (SPRED), not only enforces diffeomorphism, but additionally penalises linear, planar and volumetric changes. Application of SPRED to the high quality NIREP dataset produced results whose quality matches that of established registration methods. The resulting deformations show significantly more plausible Jacobian distributions, both in terms of spatial locality and intensity. Future work will look to extend SPRED to include variable spatial priors, allowing different brain regions freedom to deform by varying amounts.

4688
Computer 145
Hadamard Encoding Compared with Fourier Encoding in Three-dimensional (3D) Functional MRI
Seul Lee1 and Gary Glover2

1Department of Electrical Engineering, Stanford University, Stanford, CA, United States, 2Department of Radiology, Stanford University, Stanford, CA, United States

Three-dimensional (3D) acquisition is beneficial for functional MRI (fMRI) compared to two-dimensional (2D) acquisition since it can provide higher spatial resolution, resulting from potentially higher temporal signal-to-noise ratio (tSNR) and thinner slices. However, 3D has higher physiological noise due to higher signal at the center of k-space, which results in lower tSNR. The number of slices can be decreased to reduce physiological noise. However, a small number of slices in Fourier encoding results in Gibbs ringing. In this study, we show that 3D Hadamard acquisition avoids Gibbs artifacts while increasing SNR compared with conventional 2D and 3D methods.

4689
Computer 146
Myelin Water Fraction Maps with improved Fit to Noise using TGV and conventional filters
René Schranzer1,2, Günther Grabner1, Alexander Weber3, Kristian Bredies4, Gernot Reishofer5, and Alexander Rauscher3

1Department of Radiologic Technology, Carinthia University of Applied Sciences, Klagenfurt, Austria, 2Department of Engineering, University of Applied Sciences Wiener Neustadt, Wiener Neustadt, Austria, 3UBC MRI Research Centre, University of British Columbia, Vancouver, BC, Canada, 4Institute for Mathematics and Scientific Computing, University of Graz, Graz, Austria, 5Department of Radiology, Medical University of Graz, Graz, Austria

Myelin Water Imaging is the technique of choice to measure myelination changes in healthy and abnormal situations in the brain. However, calculation of myelin water fraction (MWF) maps is challenging due to the low signal-to-noise ratio in the acquired data. Here, we demonstrate different filter methods, such as TGV, Gaussian and Wiener to overcome this problem. 3D GRASE images filtered with all three methods show significant enhanced fit-to-noise (FNR) values compared to unfiltered, while TGV preserves sharper edges and detailed structures. Finally, noise reduction and thus more reliable MWF maps can lead to certain advantages in the field of MS.

4690
Computer 147
MRI guided hierarchical sectioning and stitching of brain blocks for alignment of digitized histology to corresponding MR images
Sethu K. Boopathy Jegathambal1,2,3, Kelvin Mok1,2,3, David A. Rudko1,2,3,4, and Amir Shmuel1,2,3,4

1Department of Biomedical Engineering, McGill University, Montreal, QC, Canada, 2McConnell Brain Imaging Center, McGill University, Montreal, QC, Canada, 3Montreal Neurological Institute and Hospital, McGill University, Montreal, QC, Canada, 4Department of Neurology and Neurosurgery, McGill University, Montreal, QC, Canada

Ex-vivo MRI of brain tissue can provide important morphological and microstructural information. MR images can also serve as undistorted references for reconstruction of digitized histology, immunohistochemistry or clearing techniques. However, due to space constraints imposed by small animal scanners, imaging a whole human brain generally requires a large bore scanner with limited gradient system performance and constraints on achievable spatial resolution. Importantly, large histology sections cut from large brains are prone to distortions, tears and folding. Here we present a method for MRI guided, hierarchical sectioning and stitching of brain blocks for alignment of digitized processed tissue to corresponding MRI volumes.

4691
Computer 148
Simultaneous Acquisition of Orthogonal Plane Cine Imaging and Isotropic 4D-MRI Using Super-Resolution
Nikolai J Mickevicius1 and Eric S Paulson1

1Radiation Oncology, Medical College of Wisconsin, Milwaukee, WI, United States

The super-resolution-based, isotropic 4D-MRI, and orthogonal cine imaging pulse sequence (SR-4D-SOPI) was developed and evaluated in this work. Cine imaging was acquired at more than 3 frames per second in sagittal and coronal planes simultaneously while also acquiring two orthogonal 4D-MRI volumes. The volumes were combined using super-resolution methods to create an isotropic 4D-MRI to be used for dose reconstruction following each fraction of abdominal or thoracic MR-guided radiation therapy.

4692
Computer 149
Multispectral nonlocal means filters incorporating rotations and reflections for improved noise reduction with edge preservation in magnetic resonance imaging
Nikkita Khattar1, Mustapha Bouhrara1, and Richard G. Spencer1

1Magnetic Resonance Imaging and Spectroscopy Section, National Institute on Aging, National Institutes of Health, Baltimore, MD, United States

Image denoising is used extensively for MR image post-processing. The nonlocal means (NLM) filter shows excellent noise reduction while preserving detail. NLM takes advantage of the structural redundancy in MR images by comparing local neighborhoods of voxels throughout the image, and estimating the intensity of an index voxel to be denoised through a weighted average of voxel intensities. However, this excludes patches that may be similar except for rotation or reflection, and therefore does not make full use of image redundancy.  We introduce a multispectral implementation of NLM incorporating rotations and reflections, finding improved performance compared to conventional non-multispectral filtering.

4693
Computer 150
Effects of Image Sharpening on the Accuracy of quantitative-MRI (qMRI) Maps.
Ryan McNaughton1, Mina Botros1, Xin Zhang1, and Hernan Jara1

1Boston University, Boston, MA, United States

Purpose: To study the effects of image sharpening and low spatial frequency removal on the quality of qMRI maps of T1, T2, and proton density (PD). Methods: Previously developed qMRI algorithms, augmented with specialized image filters, were tested with a gel based phantom containing three distinct solutions of variable gadolinium, sucrose, and agarose concentrations. Results: Images were successfully sharpened without significantly effecting pixel values of T1 and T2 weighted maps, while removing PD map spatial artifacts in the gadolinium vials. Conclusion: Unsharp masking and spatial flattening algorithms are effective methods for enhancing qMRI quality toward generating more accurate Synthetic-MRI maps.


Machine Learning for Image Reconstruction: Breakthroughs

Exhibition Hall
Thursday 9:15 - 10:15
 Acquisition, Reconstruction & Analysis

4694
Computer 151
EPI artifacts reduction using deep learning
Christophe Schülke1, Karsten Sommer1, and Peter Börnert1

1Philips Research, Hamburg, Germany

The inherent speed of EPI is penalized by the calibration prescans necessary to suppress N/2 ghosts. Here, we propose a deep neural network with a novel architecture that suppresses N/2 ghosts in a post-processing step starting from magnitude images, thereby eliminating the necessity of a prescan. The proposed network achieves better results than more classical networks of the same size by taking into account the N/2 structure of ghosts. The network architecture could easily be adapted to also correct for ghosts of higher order in multishot EPI.

4695
Computer 152
Convolutional Neural Networks with Aliasing Layers for Correcting Parallel Imaging and EPI Ghost Artifacts
Hidenori Takeshima1

1Advanced Technology Research Department, Research and Development Center, Canon Medical Systems Corporation, Yokohama, Japan

The author proposes a new layer named aliasing layer (AL) for effectively correcting MR-specific aliasing artifacts using convolutional neural networks. In MR images acquired using parallel imaging (PI) and/or echo-planar imaging (EPI), the locations of aliasing artifacts and/or N/2 ghost artifacts can be analytically calculated. The AL preprocesses MR images by moving the calculated locations to the locations accessible through summations over all channels in a convolution layer. The experimental results demonstrate that the correction method using the proposed AL could effectively remove PI aliasing and EPI ghosting artifacts.

4696
Computer 153
A Validation Approach for Imperfect Training Data Fidelity using Signal + Artifact + Noise-based Neural Net (SAN3)-derived Directionalized Streaking Removal
Nanyque A Boyd1, Yudai Suzuki1,2, Amit R Patel3, Jacob P Goes1, Marcella K Vaicik1, Satoru Hayamizu2, Satoshi Tamura2, and Keigo Kawaji1,3

1Biomedical Engineering, Illinois Institute of Technology, Chicago, IL, United States, 2Electrical, Electronics, and Information Engineering, Gifu University, Gifu City, Japan, 3Medicine, The University of Chicago, Chicago, IL, United States

While Deep Neural Network (DNN)-based sub-Nyquist reconstruction approaches are well-suited for high-fidelity static imaging targets such as the brain, temporally constrained (i.e. dynamic) sequences may potentially be ill-suited for DNN as these would often embed unresolved MR artifacts into the Training Data. Here, we describe an assessment approach for a generalizable DNN-based dynamic MRI reconstruction method that outputs such artifacts as characterizable and filterable streaks. This work further validates the DNN-model coding process to ensure the desired artifact/noise properties into the DNN output. Using Fourier properties, we demonstrate such validation of streaking directionalization using DNN.

4697
Computer 154
Comparison of Quality Assessment Methods for Deep-Learning-Based MR Image Reconstruction
Mohammadhassan Izady Yazdanabady 1,2, Kyoko Fujimoto3, Benjamin Paul Berman3, Matthew S Rosen4,5,6, Neha Koonjoo4,5, Bo Zhu4,5,6, Christian George Graff3, and Aria Pezeshk3

1School of Computing, Informatics, and Decision Systems Engineering, Arizona State Univerisity, Tempe, AZ, United States, 2Department of Neurosurgery, Barrow Neurological Institute, St. Joseph’s Hospital and Medical Center, Phoenix, AZ, United States, 3Center for Devices and Radiological Health, US Food and Drug Administration, Silver Spring, MD, United States, 4A. A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States, 5Harvard Medical School, Boston, MA, United States, 6Department of Physics, Harvard University, Cambridge, MA, United States

The proper methodology to perform rigorous quantitative task-based assessment of image quality for deep learning based MR reconstruction methods has not been devised yet. In this study we reconstructed T1-weighted brain images using neural networks trained with five different datasets, and explored the consistency and relationship between rankings of image quality using three different assessment metrics and FreeSurfer-based quantitative analysis. Our study indicates that assessment of image quality for a data-driven reconstruction algorithm may require several types of analysis including using different image quality assessment metrics and their agreement with clinically relevant tasks.

4698
Computer 155
Conditional generative adversarial network for three-dimensional rigid-body motion correction in MRI
Patricia Johnson1,2 and Maria Drangova1,2

1Robarts Research Institute, Western University, London, ON, Canada, 2Department of Medical Biophysics, Western University, London, ON, Canada

In this work we present a deep learning solution for motion correction in brain MRI; specifically we approach motion correction as an image synthesis problem. Motion is simulated in previously acquired brain images; the image pairs (corrupted + original) are used to train a conditional generative adversarial network (cGAN), referred to as MoCo-cGAN, to predict artefact-free images from motion-corrupted data. We also demonstrate transfer learning, where the network is fine-tuned to apply motion correction to images with a different contrast. The trained MoCo-cGAN successfully performed motion correction on brain images with simulated motion. All predicted images were quantitatively improved, and significant artefact suppression was observed. 


4699
Computer 156
Volumetric real-time imaging with deep-learning reconstruction
Jiahao Lin1,2, Fadil Ali1, and Kyunghyun Sung1

1Radiology, University of California, Los Angeles, Los Angeles, CA, United States, 2Electrical and Computer Engineering, University of California, Los Angeles, Los Angeles, CA, United States

We propose a deep-learning reconstruction pipeline for 3D real-time imaging. We use a 3D golden-angle GRE sequence, and a deep-learning network based reconstruction. Gadgetron framework is used for real-time pipelining. Using 320 images in total, our network is trained with decaying data fidelity update, and deployed without it. Dilated convolution and skip concatenation improve the image quality. We achieved a Matrix size of 192x192x8 pixels, a temporal resolution of 889ms, a reconstruction time of 300-350ms, and our image quality is comparable to iGRASP.

4700
Computer 157
Diffusion-weighted MR Image Reconstruction using Automated Transform by Manifold Approximation (AUTOMAP) on Human Brains
Koonjoo Neha1,2,3, Bo Zhu1,2,3, Matthew Christensen1,2, John E. Kirsch1,2, Bragi Sveinsson1,2,3, and Matthew S Rosen1,2,3

1A.A Martinos Biomedical Imaging Center / MGH, Charlestown, MA, United States, 2Harvard Medical School, Boston, MA, United States, 3Department of Physics, Harvard University, Cambridge, MA, United States

Low intrinsic Signal-to-Noise Ratio (SNR) in diffusion-weighted (DW) images are recurrent issues especially at high b-values. Here, we apply the deep neural network image reconstruction technique, AUTOMAP (Automated Transform by Manifold Approximation) to in-vivo diffusion-weighted MR data acquired at 1.5 T with varying b-values. In addition, apparent diffusion coefficient (ADC) maps were assessed. We also compared the reconstruction of the images using two different training corpura. The results for AUTOMAP reconstruction showed a significant increase in SNR.

4701
Computer 158
Deep Learning Based Adaptive Noise Reduction in Multi-Contrast MR Images
Kensuke Shinoda1, Kenzo Isogawa2, Masahito Nambu1, Yuichi Yamashita1, Mika Kitajima3, Hiroyuki Uetani3, and Yasuyuki Yamashita3

1MRI System Division, Canon Medical Systems Corporation, Otawara-shi, Japan, 2Corporate Research and Development Center, Toshiba Corporation, Kawasaki-shi, Japan, 3Department of Diagnostic Radiology, Faculty of Life Sciences, Kumamoto University, Kumamoto-shi, Japan

We have proposed a deep learning-based approach for MR image denoising that can adapt to the input noise power. We compare the performance of the proposed denoise approach with Deep Learning based Reconstruction (dDLR) method with state-of-the art image denoising method called Block-matching and 3D filtering method (BM3D) on multiple contrast MR images. Our experiments demonstrate that the proposed method outperforms the state-of-the art BM3D image denoising method.

4702
Computer 159
Deep Partial Fourier Reconstruction
Alexander R Toews1,2, Marcus T Alley2, Shreyas S Vasanawala2, Brian A Hargreaves1,2,3, and Joseph Y Cheng2

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Radiology, Stanford University, Stanford, CA, United States, 3Bioengineering, Stanford University, Stanford, CA, United States

Standard methods for partial Fourier (PF) reconstruction do not perform well in the presence of significant phase variations. In this study, we propose a deep-learning-based approach for PF reconstruction (DPFR) to mitigate this issue. We compare DPFR results against standard methods (Homodyne, POCS) for in vivo images of the foot, leg, and abdomen. We demonstrate that DPFR achieves superior reconstruction quality, especially near phase boundaries, across a range of partial sampling parameters. Ultimately this may extend the applicability of partial Fourier reconstruction to instances where it is not commonly used.

4703
Computer 160
Multi-supervised Learning in Cross-domain Networks for Cardiac Imaging
Ziwen Ke1,2, Shanshan Wang2, Huitao Cheng1,2, Leslie Ying3, Xin Liu2, Hairong Zheng2, and Dong Liang1,2

1Research center for Medical AI, Shenzhen Institutes of Advanced Technology, Shenzhen, China, 2Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Shenzhen, China, 3Department of Biomedical Engineering and Department of Electrical Engineering, The State University of New York, Buffalo, NY, United States

Dynamic MR image reconstruction from incomplete k-space data is an important technique for reducing its scan time. Deep learning has shown great potential in assisting this task. Nevertheless, most frameworks only adopt a final loss for network training and the intermediate results generated during the network forward pass haven't been considered for the network training.  This work proposes a multi-supervised learning strategy, which constrains the frequency domain information and reconstruction results at different levels. Improved reconstruction results have been achieved with the proposed strategy. 

4704
Computer 161
Ultra-low-dose Amyloid PET/MRI Reconstruction by Generative Adversarial Network
Jiahong Ouyang1, Kevin T. Chen2, Enhao Gong2, John Pauly2, and Greg Zaharchuk2

1Carnegie Mellon University, Pittsburgh, PA, United States, 2Stanford University, Stanford, CA, United States

Amyloid PET is widely used in the early diagnosis of dementia. However, the injection of the radiotracer will lead to radiation exposure to the subject. We proposed a novel method based on Generative Adversarial Network (GAN) with perceptual loss to achieve diagnostic image quality PET images using ultra-low-dose PET images with or without additional MR contrasts as inputs.

4705
Computer 162
Improved TWIST Imaging using k-Space Deep Learning
Eunju Cha1, Eung Yeop Kim2, and Jong Chul Ye1

1KAIST, Daejeon, Korea, Republic of, 2Gachon University Gil Medical Center, Incheon, Korea, Republic of

Time-resolved angiography with interleaved stochastic trajectories(TWIST) has been widely used for dynamic contrast enhanced (DCE) MRI. To achieve highly accelerated acquisitions for improved temporal and spatial resolution, the high frequency region is randomly sub-sampled at each time frame. Therefore, the periphery of the k-space data from multiple time frames are combined to obtain the uniformly sub-sampled k-space data so that the temporal resolution of TWIST is limited. The purpose of this research is to improve the temporal resolution of TWIST by reducing the view-sharing. Furthermore, we proposed the algorithm that can reconstruct the imagesat various number of view sharing using k-space deep learning.

4706
Computer 163
Contrast Transfer Learning for Reconstruction of Undersampled Dynamic Contrast-Enhanced MRI
Li Feng1, Fang Liu2, Lihua Chen3,4, and Ricardo Otazo1,5

1Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, United States, 2Department of Radiology, University of Wisconsin-Madison, Madison, WI, United States, 3Department of Radiology, Southwest Hospital, Chongqing, China, 4Department of Radiology, PLA 101st Hospital, Wuxi, Jiangsu, China, 5Department of Radiology, Memorial Sloan Kettering Cancer Center, New York, NY, United States

The application of deep learning for reconstruction of dynamic contrast-enhanced MRI presents significant challenges caused by the rapid passage of the contrast agent, which makes it difficult to acquire fully-sampled images to train a neural network. This work proposes to use images from a delayed contrast phase, where contrast changes are in a relatively steady state, for training, and to apply the trained neural network for reconstruction of undersampled data acquired in other contrast phases. The proposed contrast transfer learning reconstruction was trained on 55 post-contrast liver cases and tested on a first-pass liver DCE-MR acquisition.

4707
Computer 164
Development of a deep learning method for phase unwrapping MR images
Kanghyun Ryu1, Sung-Min Gho2, Yoonho Nam3, Kevin Koch4, and Dong-Hyun Kim1

1Department of Electrical and Electronic Engineering, Yonsei University, Seoul, Korea, Republic of, 2MR Clinical research and Development, GE Healthcare, Seoul, Korea, Republic of, 3Seoul St.Mary's Hospital, The Catholic University of korea, Seoul, Korea, Republic of, 4Department of Radiology, Medical College of Wisconsin, Milwaukee, WI, United States

MRI phase images are increasingly used for susceptibility mapping and distortion correction in function and diffusion MRI. However, acquired values of phase maps are wrapped between [-π π ] and require an additional phase unwrapping process. Here we developed a novel deep learning method that can learn the transformation between the wrapped phase images and the corresponding unwrapped phase images. The method was tested for numerical simulations and on actual MR images.

4708
Computer 165
Ultrafast 3D Partial Fourier Reconstruction with Well-Preserved Phase using DNN
Guangliang Ju1, Aiqi Sun2, Mingliang Chen2, Yu Wang2, Dong Han1, and Feng Huang2

1Neusoft Medical Systems, Shenyang, China, 2Neusoft Medical Systems, Shanghai, China

Partial Fourier (PF) is a widely used fast imaging scheme. Since phase information is crucial in many applications, such as SWI, it is necessary that PF can preserve phase well. Many PF methods cannot preserve phase well especially at locations with rapid phase change. DPA is a method can recover both magnitude and phase well, but suffers from low speed for two-directional PF acquisition. Considering recent advances in deep learning, we proposed a DNN-based framework for two-directional PF reconstruction. Preliminary experiments demonstrate that the proposed method is almost 50 times faster while restores magnitude and SWI even better than DPA.

4709
Computer 166
Accelerated MR image reconstruction using iterative feasible set projection
Doohee Lee1,2, Jaeyeon Yoon1,2, Jingyu Ko1,2, Jingu Lee1,2, Yoonho Nam3,4, and Jongho Lee1

1Seoul National University, Seoul, Korea, Republic of, 2AIRS medical, Seoul, Korea, Republic of, 3Department of Radiology, Seoul St. Mary’s Hospital, Seoul, Korea, Republic of, 4College of Medicine, The Catholic University of Korea, Seoul, Korea, Republic of

 We proposed a new deep learning architecture for the reconstruction of highly undersampled data. The new architecture combines an iterative generative adversarial network (GAN) with a shared discriminator and interacts with data consistency blocks. The algorithm was applied to accelerate the data acquisition of the routine clinical protocols, particularly 2D Cartesian sampling sequences. The new method was tested to explore generalizability of the algorithm in in-vivo data under various conditions (difference pulse sequences, organs, coil types, sites, and health condition).

4710
Computer 167
Rapid Image Reconstruction of Single-Shot Coronary Quiescent-Interval Slice-Selective (QISS) MRA and Late Gadolinium-Enhanced MRI using Deep Learning
Daming Shen1,2, Hassan Haji-Valizadeh1,2, and Daniel Kim1,2

1Biomedical Engineering, Northwestern University, Evanston, IL, United States, 2Radiology, Northwestern University, Chicago, IL, United States

While compressed sensing (CS) enables highly-accelerated cardiac MRI acquisitions, its lengthy image reconstruction may limit clinical translation. Deep learning (DL) is capable reconstructing undersampled images with clinically acceptable reconstruction times. The purpose of this study was to build, train, and validate a deep learning framework for rapidly reconstructing highly-accelerated cardiac MR images, where CS reconstructed images are used as reference. 

4711
Computer 168
Enforcing Structural Similarity in Deep Learning MR Image Reconstruction
Kamlesh Pawar1,2, Zhaolin Chen1,3, N Jon Shah4, and Gary F Egan1,2

1Monash Biomedical Imaging, Monash University, Melbourne, Australia, 2School of Psychology, Monash University, Melbourne, Australia, 3Electrical and Computer System Engineering, Monash University, Melbourne, Australia, 4Institute of Medicine, Research Centre Juelich, Juelich, Australia

Deep Learning (DL) MR image reconstruction from undersampled data involves minimization of a loss function. The loss function to be minimized drives the DL training process and thus determine the features learned. Usually, a loss function such as mean squared error or mean absolute error is used as the similarity metric. Minimizing such loss function may not always predict visually pleasing images required by the radiologist. In order to predict visually appealing MR images in this work, we propose to use the difference of structural similarity as a regularizer along with the mean squared loss. 

4712
Computer 169
Edge-enhanced Loss Constraint for Deep Learning Based Dynamic MR Imaging
Shanshan Wang1, Ziwen Ke1,2, Huitao Cheng1,2, Leslie Ying3, Xin Liu1, Hairong Zheng1, and Dong Liang1,2

1Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institutes of Advanced Technology, Shenzhen, China, 2Research center for Medical AI, Shenzhen Institutes of Advanced Technology, Shenzhen, China, 3Department of Biomedical Engineering and Department of Electrical Engineering, The State University of New York, Buffalo, NY, United States

Cardiac magnetic resonance (MR) imaging provides a powerful imaging tool for clinical diagnosis. However, due to the constraints of magnetic resonance (MR) physics and reconstruction algorithms, dynamic MR imaging takes a long time to scan. Recently, deep learning has achieved preliminary success in MR reconstruction. Compared with the classical iterative optimization algorithms, the deep learning based methods can get improved reconstruction results in shorter time. However, most current deep convolutional neural network (CNN) based methods use mean square error (MSE) as the loss function, which might be a reason for image smooth in the reconstruction. In this work, we  propose to employ edge-enhanced constraint for loss function and explore different types of total variation on network training. Encouraging performances have been achieved. 

4713
Computer 170
Exploration on deep-learning based sorting of k-space data for ECG-free cardiac cine-MRI
Seb Harrevelt1, Tim Leiner2, J.P.W. Pluim3, and A.J.E. Raaijmakers2

1TU Eindhoven, Delft, Netherlands, 2UMC Utrecht, Utrecht, Netherlands, 3TU Eindhoven, Eindhoven, Netherlands

Cine-cardiac MRI reconstruction relies on the ECG signal to sort k-space data. However, ECG triggering comes with disadvantages among which increased setup time. Here we suggest an alternative method of sorting cine MRI k-space data using deep-learning.  An explorative study has been performed using an encoder-decoder network with Sinkhorn layer to sort k-space data that was randomly disordered in one spatial dimension. Good reconstructions were obtained using a group size of 8 or more k-space lines during randomization. These results hold promise for subsequent application in the time dimension.

4714
Computer 171
Complex-Valued Convolutional Neural Networks for MRI Reconstruction
Elizabeth Cole1, John Pauly1, Shreyas Vasanawala2, and Joseph Cheng2

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Radiology, Stanford University, Stanford, CA, United States

To improve MRI reconstruction accuracy, we propose various complex-valued frameworks for reconstructions using convolutional neural networks. By introducing complex-valued convolution and activation functions, we improve reconstruction of our subsampled images and achieve competitive results compared to the real-valued counterpart of our model.


4715
Computer 172
Deep Inception Residual Network (DIRN) for Reconstruction of Undersampled Brain MR Image
Sekeun Kim1, Yeonggul Jang2, Hackjoon Shim3, and Hyukjae Chang4

1Graduate program in Biomedical Engineering The Graduate School, Yonsei University, Seoul, Korea, Republic of, 2Brain Korea 21 PLUS Project for Medical Science, Yonsei University, seoul, Korea, Republic of, 3Yonsei-Cedars-Sinai Integrative Cardiac Imaging Research Center, seoul, Korea, Republic of, 4Division of Cardiology, Severance Cardiovascular Hospital, Yonsei University, College of Medicine, seoul, Korea, Republic of

Acquiring the full-sampling k-space magnetic resonance imaging (MRI) data for detailed anatomical information is ideal. We propose the Deep Inception Residual Network (DIRN) based on a deep convolutional neural network (DCNN) consisting of inception blocks and residual blocks for the reconstruction of the MR image from undersampled k-space data. The experimental results on an ADNI dataset demonstrate that DIRN is appropriate for a reconstruction of brain MR images.

4716
Computer 173
g-factor attention model for deep neural network powered parallel imaging: gANN
Jaeyeon Yoon1,2, Doohee Lee1,2, Jingyu Ko1,2, Jingu Lee1, Yoonho Nam3,4, and Jongho Lee1

1Seoul National University, Seoul, Korea, Republic of, 2AIRS medical, Seoul, Korea, Republic of, 3Department of Radiology, Seoul St. Mary’s Hospital, Seoul, Korea, Republic of, 4College of Medicine, The Catholic University of Korea, Seoul, Korea, Republic of

In this study, we proposed a new concept of an attention model for deep neural network based parallel imaging. We utilized g-factor maps to inform the neural network about the location containing high possibility of aliasing artifact. Also the proposed network used sensitivity maps and acquired k-space data to ensure the data consistency. Since the g-factor attention deep neural network considered both multi-channel information and spatially variant aliasing condition, our proposed network successfully removed aliasing artifacts up to factor 6 in uniform under-sampling and showed high performance when compared to conventional parallel imaging methods.

4717
Computer 174
Accelerating high resolution DWI via deep learning
Da Zou1, Ruibo Song1, Dong Han1, and Feng Huang1

1Neusoft Medical Systems, Shenyang, China

The conventional multi-shot diffusion weighted imaging (DWI) techniques, such as MUSE, have not been widely adopted clinically due to long scan time. In this study, an accelerated multi-shot DWI method based on deep learning is proposed. By learning a fully convolutional neural network to enhance DWI images, more structural details and less noise can be achieved, especially when with fewer shots or NSA (Number of Signal Average), in the meantime the reconstruction time can be reduced by over 200 times. It means the proposed approach reduces the scan and reconstruction time dramatically while keeping high quality of the images, which makes it a potential technique for high resolution multi-shot DWI in routine clinical study.

4718
Computer 175
A Generative Adversarial Network with a Progressively Growing Training Strategy for MRI Dataset Augmentation
Vahid K Ghodrati1,2, Haotian An3, Zihao Xiong3, Jiaxin Shao1, Mark Bydder1, and Peng Hu1

1Radiology, University of California Los Angeles, Los Angeles, CA, United States, 2Biomedical Physics Inter-Departmental Graduate Program, University of California Los Angeles, Los Angeles, CA, United States, 3Department of Electronic Engineering, Tsinghua University, Beijing, China

For medical imaging applications, it is not straightforward to create a large database due to high costs associated with acquiring the data, patent privacy issues, and challenges in pooling data from multiple medical institutions. Generating high-resolution medical images from the latent noise vector could potentially mitigate training data size issues in applying DNN to medical imaging. This could facilitate objective comparisons between the different machine learning algorithms in medical imaging. In this study, progressive growing strategy is considered to train the GAN stably and generate super resolution brain datasets from noise vector.  


Segmentation 1

Exhibition Hall
Thursday 13:45 - 14:45
 Acquisition, Reconstruction & Analysis

4719
Computer 1
Automated femoro-tibial cartilage segmentation of OA patients with and without bone abnormality
Rafeek Thaha1, Sandeep Panwar Jogi1,2, Sriram Rajan3, Vidur Mahajan3, Vasantha K Venugopal3, Amit Mehndiratta1,4, Anup Singh1,4, Dharmesh Singh1, and Neha Vats5

1Centre for Biomedical Engineering, Indian Institute of Technology, New Delhi, India, 2Biomedical Engineering, ASET, Amity University Haryana, Gurgaon, India, 3Mahajan Imaging Centre, New Delhi, India, 4Department of Biomedical Engineering, All India Institute of Medical Sciences, New Delhi, India, 5National Institute of Technology, Kurukshetra, India

The study of knee cartilage under subchondral abnormality is important in osteoarthritis (OA) progression studies. However, cartilage segmentation for patients with Bone-Marrow-Edema (BME) lesion, particularly using radial-search based approach, is erroneous. In this study, a framework for automatic segmentation of femoro-tibial cartilage of OA patients with and without bone abnormality, based on modified radial-search approach and T2-map values is developed. A 2D projected view of T2-map and thickness values of cartilage was generated. Proposed method was successfully applied on 23 MRI patient data. Dice-coefficient for cartilage segmentation was ~82% for OA patients with and without BME lesions.  

4720
Computer 2
Performant summative 3D rendering of voxel-wise MRF segmentation data
Andrew Dupuis1,2, Debra McGivney3, Rasim Boyacioglu3, Dan Ma3, Anagha Deshmane4, and Mark A Griswold1,2,3

1Biomedical Engineering, Case Western Reserve University, Cleveland, OH, United States, 2Interactive Commons, Case Western Reserve University, Cleveland, OH, United States, 3Department of Radiology, School of Medicine, Case Western Reserve University, Cleveland, OH, United States, 4Max Planck Institute for Biological Cybernetics, Tübingen, Germany

Visualization of Magnetic Resonance Fingerprinting segmented data presents significant difficulty because of the abstraction from the usual appearance and contrast of MR images. We present a method of rendering any probability-based tissue fraction partial volume ROIs in three dimensions using additive voxelized volumetric rendering as a form of segmentation. Datasets consist of n groups of segmented maps with each voxel representing the probability of a given tissue converted into 3D textures usable by the GPU to perform raymarched additive rendering. This allows for different tissue classifications within the dataset to be faded in and out with minimal human involvement.

4721
Computer 3
Segmentation and probabilistic tractography of GPi, GPe, STN and RN using Lead-DBS and FSL
Jae-Hyuk Shim1 and Hyeon-Man Baek2

1Department of Health Sciences and Technology, GAIHST, Gachon University, Incheon, Korea, Republic of, 2Gachon University, Incheon, Korea, Republic of

Lead-DBS toolbox is used to segment globus pallidus internal, globus pallidus external, subthalamic nucleus and red nucleus, all of which are structures not automatically segmented by popular toolboxes such as FSL and Freesurfer. In addition, FSL's diffusion toolbox was used to generate probabilistic tractography between each segmented structure as well as compare the level of connectivity between each segmented structure.

4722
Computer 4
Deep learning segmentation (AxonDeepSeg) to generate axonal-property map from ex vivo human optic chiasm using light microscopy
Thibault Tabarin1, Maria Morozova2,3, Carsten Jaeger2, Henriette Rush3, Markus Morawski3, Stefan Geyer2, and Siawoosh Mohammadi1

1Department of Neurophysics, Medical center Hamburg-Eppendorf, Hamburg, Germany, 2Department of Neurophysics, Max Planck Institute for Human cognitive and Brain Sciences, Leipzig, Germany, 3Paul Flechsig Institut of Brain Research, University Leipzig, Leipzig, Germany

Development of in-vivo histology using MRI needs validation strategies with gold standard methods. Ex-vivo histology combined with microscopy could become such a strategy; however, for comparing larger field-of-views automatic segmentation of axons and myelin will be required. State-of-the-art segmentation has recently involved deep learning (DL). In this work, we investigated the recently published AxonDeepSeg deep learning algorithm (ADS). We successful applied ADS on light microscopy images of an optical chiasm sample, improved the segmentation of myelin to access the full properties of individual fibers, and finally created microstructural maps such as the histology g-ratio map.

4723
Computer 5
Segment Unannotated MR Image Dataset using Joint Image Translation and Segmentation Adversarial Network
Fang Liu1

1Radiology, University of Wisconsin-Madison, Madison, WI, United States

The purpose of our study was to develop and evaluate a generalized CNN-based method for fully-automated segmentation of different MR image datasets using a single set of annotated training data. A technique called cycle-consistent generative adversarial network (CycleGAN) is applied as the core of the proposed method to perform image-to-image translation between MR image datasets with different tissue contrasts. A joint segmentation network is incorporated into the adversarial network to obtain additional segmentation functionality. The proposed method was evaluated for segmenting bone and cartilage on two clinical knee MR image datasets acquired at our institution using only a single set of annotated data from a publicly available knee MR image dataset. The new technique may further improve the applicability and efficiency of CNN-based segmentation of medical images while eliminating the need for large amounts of annotated training data.

4724
Computer 6
Automated Segmentation of Substantia Nigra in Neuromelanin-Sensitive Magnetic Resonance Imaging
Touseef Ahmad Qureshi1, Cody Lynch1, Elliot Hogg2, Tina Wu3, Michele Tagliati3, Debiao Li1, and Zhaoyang Fan1

1Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, CA, United States, 2Department of Neurology, Cedars-Sinai Medical Center, Los Angeles, CA, United States, 3Cedars-Sinai Medical Center, Los Angeles, CA, United States

Accurate segmentation of Substantia Nigra (SN) in Neuromelanin-Sensitive MRI (NM-MRI) is a prerequisite for efficient quantification and evaluation of severity of Parkinson disease. We present a fully automated algorithm for localization and segmentation of SN in NM-MRI. The localization algorithm uses a new specialized template matching model consisting of a resizable cardioid plane. The segmentation of SN is performed using freeform active contour segmentation model. The system is tested on 19 NM-MRI scans (10 healthy volunteers and 9 patients with Parkinson disease), acquired using 3T MRI system. The success rate for localization is 98.2%, whilst dice coefficient for segmentation reaches 0.89.

4725
Computer 7
Rapid virtually automated technique for renal corticomedullary segmentation from volumetric arterial phase imaging: Initial experience
Kane Nicholls1, Julia Williams1, Lucy McKenna2, Julie Smith2, Emma Hornsey2, Elif Ekinci2, Leonid Churilov3, Henry Rusinek4, Artem Mikheev5, and Ruth P Lim1

1Radiology, Austin Health, Heidelberg, Australia, 2Austin Health, Heidelberg, Australia, 3Florey Institute of Neuroscience and Mental Health, Heidelberg, Australia, 4Radiology, New York University, New York, NY, United States, 5New York University, New York, NY, United States

Efficient, reproducible and accurate corticomedullary renal segmentation is challenging but important for MR renography and disease monitoring. We assessed segmentation time, reproducibility and accuracy of a virtually automated (VA) approach (<5 second user interaction), compared to gold standard (GS) manual segmentation. Segmentation time per subject (n=11) was 78.6±7.0s for VA and 60-120min for GS. VA intra- and inter-rater agreement was near perfect for cortex, medullary and whole kidney segmentation (concordance correlation coefficient all ≥0.99), with excellent concordance with GS segmentation (CCC all >0.80). VA is a rapid, accurate and highly reproducible corticomedullary segmentation tool which has promising clinical potential.

4726
Computer 8
Performance of Automatic Cerebral Arterial Segmentation of MRA Images Improves in Patients with Anemia and Sickle Cell Disease Compared with Healthy Volunteers.
Alexander Saunders1, John C. Wood2, and Matthew Borzage3,4

1Viterbi School of Engineering, University of Southern California, Los Angeles, CA, United States, 2Division of Cardiology, Children's Hospital Los Angeles, Los Angeles, CA, United States, 3Division of Neonatology, Children's Hospital Los Angeles, Los Angeles, CA, United States, 4Department of Pediatrics, Keck School of Medicine, University of Southern California, Los Angeles, CA, United States

Sickle cell disease (SCD) and chronic anemia cause morphological abnormalities in the cerebral arterial vasculature that are observable using time-of-flight magnetic resonance angiography (MRA). We seek to evaluate the accuracy of automatic vessel segmentation algorithms in extracting vessel data from these images for further analysis. Five segmentation algorithms were applied to three MRA images (one control, one anemic, and one SCD patient) and performance was measured against manually segmented ground truth data. We found that automatic segmentation performs better in anemic and SCD patients over healthy controls.

4727
Computer 9
Automated intervertebral disc segmentation using a two-pathway network
Fei Gao1, Shui Liu2, Xiaodong Zhang2, Jue Zhang1,3, and Xiaoying Wang2,3

1College of Engineering, Peking University, Beijing, China, 2Department of Radiology, Peking University First Hospital, Beijing, China, 3Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, China

We developed a two-pathway fully convolutional network for refined intervertebral disc segmentation. The proposed pooling free subbranch can capture more local fine-grained features. The quantitative results indicate its priority for disc segmentation.

4728
Computer 10
The deep learning lesion segmentation method nicMSlesions only needs one manually delineated subject to outperform commonly used unsupervised methods
Merlin M Weeda1, Iman Brouwer1, Marlieke L de Vos1, Myrte S de Vries1, Frederik Barkhof1,2, Petra JW Pouwels1, and Hugo Vrenken1

1Department of Radiology and Nuclear Medicine, MS Center Amsterdam, Amsterdam Neuroscience, Amsterdam UMC - location VUmc, Amsterdam, Netherlands, 2Institutes of Neurology and Healthcare Engineering UCL, London, United Kingdom

Automatic lesion segmentation is important for measurements of atrophy and lesion load in subjects with multiple sclerosis (MS). Although supervised methods perform overall better than unsupervised methods, they are not widely used since they are more labor-intensive due to the need for great amounts of manual input. Our research showed increased performance of supervised methods over unsupervised methods. In addition, when using a deep learning based supervised method, training on only one subject already outperformed the commonly used unsupervised methods. We therefore recommend using deep learning lesion segmentation methods in MS research. 

4729
Computer 11
Propagation Neural Network for cardiac segmentation
Benjamin Roussel1,2, Julien Oster2,3, and Mattias Paul Heinrich4

1Université de Lorraine, Nancy, France, 2U1254, INSERM, Nancy, France, 3Université de Lorraine, Nancy, France, Metropolitan, 4Universität zu Lübeck · Institut für Medizinische Informatik, Lübeck, Germany

  To perform a fully-automated segmentation of cardiac volumes, current Convolutional Neural Networks (CNNs) process each slice independently, not taking the depth information into consideration. Networks using 3D convolutions being memory-hungry, we propose a CNN model with a low memory demand and processing the whole volume. The network is based on propagating the redundant depth information from slice to slice. Following a 4-fold cross validation on the MICCAI/ACDC challenge dataset, our network obtained better results than a standard 2D network, improving the average DICE score of 1.7% computed over three cardiac structures (myocardium, left and right ventricle).

4730
Computer 12
Automated organ segmentation of liver and spleen in whole-body T1-weighted MR images: Transfer learning between epidemiological cohort studies
Thomas Kuestner1,2,3, Sarah Müller2, Marc Fischer2,3, Martin Schwartz2,3, Petros Martirosian2, Bin Yang3, Fritz Schick2, and Sergios Gatidis2

1School of Biomedical Engineering & Imaging Sciences, King's College London, London, United Kingdom, 2Department of Radiology, University Hospital Tübingen, Tübingen, Germany, 3Institute of Signal Processing and System Theory, University of Stuttgart, Stuttgart, Germany

Automated segmentation of organs and anatomical structures is a prerequisite for efficient analysis of MR data in large cohort studies with thousands of participants. The feasibility of deep learning approaches has been shown to provide good solutions. Since all these methods are based on supervised learning, labeled ground truth data is required which can be time- and cost-intensive to generate. This work examines the feasibility of transfer learning between similar epidemiological cohort studies to derive possibilities in reuse of labeled training data.

4731
Computer 13
Beyond Dice Coefficient: Evaluating Shape Biomarker Preservation in Neural Network Segmentations
Claudia Iriondo1,2, Valentina Pedoia1, Michael Girard3, and Sharmila Majumdar1

1Radiology and Biomedical Imaging, University of California, San Francsico, San Francisco, CA, United States, 2Bioengineering, University of California, Berkeley, Berkeley, CA, United States, 3Center for Digital Health Innovation, University of California, San Francsico, San Francisco, CA, United States

High accuracy scores in volumetric overlap metrics, such as Dice Similarity Coefficient, have not been proven to be reliable indicators of shape biomarker preservation. This study proposes a novel approach towards quantitative evaluation of segmentations from neural networks using PCA and contrastive PCA.

4732
Computer 14
Fully Automatic Learning-based Multi-Organ Segmentation(ALMO) in abdominal MRI for Radiotherapy Planning using Deep Neural Networks
Yuhua Chen1,2, Yujin Xie3, Lixia Wang1,4, Jiayu Xiao1, Zixin Deng1, Yi Lao5, Richard Tuli5, Debiao Li1, Wensha Yang5, and Zhaoyang Fan1

1Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, CA, United States, 2Biomedical Engineering, UCLA, Los Angeles, CA, United States, 3Beihang University, Beijing, China, 4Beijing Chaoyang Hospital, Capital Medical University, Beijing, China, 5Department of Radiation Oncology, Cedars-Sinai Medical Center, Los Angeles, CA, United States

Precise dose measurement is critical in radiotherapy planning, which involves accurate and fast segmentation of the organ for estimation of the region at risk. Segmentation Magnetic Resonance Imaging (MRI), as it is gaining more favor against CT in radio therapy, is new for multi-organ segmentation task. In this work, we proposed a fast, accurate, and fully automatic technique (ALMO) that reliefs the intense human labor from manual segmentation in a timing fashion. On our 51-subject dataset, our proposed method achieves an average dice score of 0.76 in the test set in seconds.

4733
Computer 15
Automatic Detection and Segmentation of Brain Metastases using Deep Learning on Multi-Modal MRI: A Multi-Center Study
Endre Grøvik1,2, Darvin Yi3, Michael Iv2, Elizabeth Tong2, Kyrre Eeg Emblem1, Line Brennhaug Nilsen1, Cathrine Saxhaug4, Kari Dolven Jacobsen5, Åslaug Helland5, Daniel Rubin3, and Greg Zaharchuk2

1Department for Diagnostic Physics, Oslo University Hospital, Oslo, Norway, 2Department of Radiology, Stanford University, Stanford, CA, United States, 3Department of Biomedical Data Science, Stanford University, Stanford, CA, United States, 4Department of Radiology and Nuclear Medicine, Oslo University Hospital, Oslo, Norway, 5Department of Oncology, Oslo University Hospital, Oslo, Norway

In recent years, many deep learning approaches have been developed and tested for automatic segmentation of gliomas. However, few studies have shown its potential for use in patients with brain metastases. Deep learning may ultimately aid radiologists in the tedious and time-consuming task of lesion segmentation. The objective of this work is to assess the clinical potential and generalizability of a deep learning technique, by training and testing a convolutional neural network for segmenting brain metastases using multi-center data. 

4734
Computer 16
3D U-Net for Automated Segmentation of the Thoracic Aorta in 4D-Flow derived 3D PC-MRA
Haben Berhane1, Michael Scott2, Joshua Robinson1, Cynthia Rigsby1, and Michael Markl2

1Lurie Childrens Hospital of Chicago, Chicago, IL, United States, 2Northwestern University, Chicago, IL, United States

We developed a 3D convolutional neural network for the automatic segmentation of the thoracic aorta in 4D Flow-derived 3D PC-MRAs. Using 100 testing datasets, we obtained an average dice score of 0.94±0.03 and an average voxel-wise accuracy of 0.99. Additionally, our algorithm is robust enough to accurately segment a wide array of aortic geometries and disease, such as bicuspid aortic value, coarctation, and interrupted aortic arches.

4735
Computer 17
Prostate and peripheral zone segmentation on multi-vendor MRIs using Deep Learning
Olmo Zavala-Romero1, Adrian L. Breto1, Nicole Gautney1, Yu-Cherng C. Chang1, Alan Dal Pra1, Mattew C Abramowitz1, Alan Pollack1, and Radka Stoyanova1

1Radiation Oncology, University of Miami, Miami, FL, United States

A Deep Learning algorithm for automatic segmentation of the prostate and its peripheral zone (PZ) is investigated across MR images from two MRI vendors. The proposed architecture is a 3D U-net that uses axial, coronal, and sagittal MRI series as input. When trained with Siemens MRI, the network achieves a Dice similarity coefficient (DSC) of .91 and .76 for the segmentation of the prostate and the PZ respectively. However, the network performs poorly on a GE dataset. Combining images from different MRI vendors is of paramount importance to pursue a universal algorithm for prostate and PZ segmentation.

4736
Computer 18
Technical Considerations for Semantic Segmentation in Magnetic Resonance Imaging using Deep Convolutional Neural Networks: A Case Study in Femoral Cartilage Segmentation
Arjun D. Desai1, Garry E. Gold1,2,3, Brian A. Hargreaves1,2,4, and Akshay S. Chaudhari1

1Radiology, Stanford University, Stanford, CA, United States, 2Bioengineering, Stanford University, Stanford, CA, United States, 3Orthopedic Surgery, Stanford University, Stanford, CA, United States, 4Electrical Engineering, Stanford University, Stanford, CA, United States

Deep convolutional neural networks (CNNs) have shown promise in challenging tissue segmentation problems in medical imaging. However, due to the large size of these networks and stochasticity of the training process, the factors affecting CNN performance are difficult to analytically model. In this study, we numerically evaluate the impact of network architecture and characteristics of training data on network performance for segmenting femoral cartilage. We show that extensive training of several common network architectures yields comparable performance and that somewhat optimal network generalizability can be achieved with limited training data.

4737
Computer 19
Conditional adversarial network for segmentation with simple loss function
Andre Maximo1 and Chitresh Bhushan2

1GE Healthcare, Rio de Janeiro, Brazil, 2GE Global Research, Niskayuna, NY, United States

Most deep-learning approaches require defining a loss function that is appropriate for the task. The choice of the loss function generally substantially affects the accuracy of the trained model and often requires hand-tuning. For example, some segmentation tasks work well with Dice loss while other work well with mean squared error (MSE). In this work we show how conditional adversarial network (cGAN) can be used to avoid defining a specialized loss function for each task and, instead, use a simple approach to achieve comparable or even superior results in context of segmentation of MRI images.

4738
Computer 20
Towards Domain-invariant Carotid Artery Lumen-wall Segmentation Using Adversarial Networks
Anna Danko1,2, Roberto Souza2,3, and Richard Frayne2,3

1Medical Sciences Graduate Program, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada, 2Seaman Family MR Research Centre, Foothills Medical Centre, Calgary, AB, Canada, 3Radiology and Clinical Neuroscience, Hotchkiss Brain Institute, University of Calgary, Calgary, AB, Canada

Magnetic resonance (MR) imaging is frequently used for carotid artery wall imaging. The capacity for multi-contrast imaging allows MR scanners to resolve the lumen and wall, as well as multiple plaque components. Combined this information can provide evidence of increased stroke risk. Quantitative analysis of carotid artery MR images regularly begins with the manual segmentation of wall and plaque. This process is time-consuming and costly, and suggests the need for automated methods. Developing a robust segmentation tool is challenging because of the domain shift due to different image contrasts and/or scanners. Here, we demonstrate that a deep learning network including an adversarial component is capable of learning domain-invariant features, thus producing a generalizable segmentation model.

4739
Computer 21
Deep convolutional neural networks for brain lesion segmentation in multiple sclerosis using clinical MRI scans
Sunny Nagam1, Glen Pridham1, and Yunyan Zhang1

1University of Calgary, Calgary, AB, Canada

Machine learning opens up a new opportunity for advancing our image pattern recognition abilities in medical imaging. In this study, we tested the potential of 3 new deep convolutional neural network-based learning methods for detecting brain MRI lesions in multiple sclerosis (MS). Using clinical scans available online from 10 patients, we found that the ResNet and SegNet achieved a promising dice score of 0.65 and 0.61 respectively, better than the generative adversarial network. Deep learning methods may be novel tools for optimal detection of brain MRI lesions, improving the management of patients with MS and similar disorders.

4740
Computer 22
Development of U-Net Breast Density Segmentation Method for Fat-Sat T1-Weighted Images Using Transfer Learning from Model for Non-Fat-Sat Images
Yang Zhang1, Jeon-Hor Chen1,2, Kai-Ting Chang1, Siwa Chan3, Huay-Ben Pan4, Jiejie Zhou5, Ouchen Wang6, Meihao Wang5, and Min-Ying Lydia Su1

1Department of Radiological Sciences, University of California, Irvine, CA, United States, 2Department of Radiology, E-Da Hospital and I-Shou University, Kaohsiung, Taiwan, 3Department of Medical Imaging, Taichung Tzu-Chi Hospital, Taichung, Taiwan, 4Department of Radiology, Kaohsiung Veterans General Hospital, Kaohsiung, Taiwan, 5Department of Radiology, The First Affiliate Hospital of Wenzhou Medical University, Wenzhou, China, 6Department of Thyroid and Breast Surgery, The First Affiliate Hospital of Wenzhou Medical University, Wenzhou, China

The U-Net deep learning is a feasible method for segmentation of breast and fibroglandular tissue on non-fat-suppressed (non-fat-sat) T1-weighted images. Whether it can work on fat-sat images, which are more commonly used for diagnosis, is studied. Three datasets were used: 126 Training, 62 Testing Set-A, and 41 Testing Set-B. The model was developed without and with transfer learning based on parameters in the previous model developed for non-fat-sat images. The results show that U-Net can also achieve a high segmentation accuracy for fat-sat images, and when training case number is small, transfer learning can help to improve accuracy.

4741
Computer 23
Complete Segmentation of Human Thigh and Calf Muscles/Tissues with Convolutional Neural Network and Partially Segmented Training Images
Chun Kit Wong1, Tian Siew Yap1,2, Serene Shi Hui Teo1, Maria Kalimeri1, and Mary Charlotte Stephenson1,2

1Clinical Imaging Research Center, National University of Singapore, Singapore, Singapore, 2AGENCY FOR SCIENCE, TECHNOLOGY AND RESEARCH (A*STAR), Singapore, Singapore

Quantitative analysis of lower extremity images typically require manual or semi-automated segmentation of regions of interest. This can be extremely time consuming. Here, we utilise DeepLearning and a database of previously segmented thigh and calf t1-weighted images to automatically segment the images into different tissue types and various muscle groups. Dice scores greater than 0.85 were achieved on average across the classes with as few as 40 training images (3D). In addition, we demonstrate a method for training the model with partially labelled images, enabling access to potentially much larger training datasets.

4742
Computer 24
An fully automatic prostate segmentation based on generative adversarial networks
Yi Zhu1, Rong Wei1, Ge Gao2, Jue Zhang1,3, and Xiaoying Wang2

1Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, China, 2Peking University First Hospital, Beijing, China, 3College of Engineering, Peking University, Beijing, China

Automatic prostate segmentation in MR images is essential in many clinical applications. Generative adversarial networks(GAN) have recently gained interests due to their promising ability in generating images which are difficult to distinguish from real images. In this paper, we propose an automatic and efficient algorithm base on GAN to segment the prostate contour and make the prostate segmentation shape more realistic. Our restult shows that the mean segmentation accuracy in test dataset is 90.3%±5.5. It indicates that the proposed strategy is feasible for segmentation of prostate MR images.


4743
Computer 25
Automatic Segmentation of Carotid Vessel Wall on GOAL-SNAP Images using SE-UNet
Yuze Li1, Haikun Qi2, Huiyu Qiao1, Hualu Han1, Xihai Zhao1, Chun Yuan1,3, and Huijun Chen1

1Center for Biomedical Imaging Research, Department of Biomedical Engineering, Tsinghua University, Beijing, China, 2School of Biomedical Engineering and Imaging Sciences, King's College, London, United Kingdom, 3Department of Radiology, University of Washington, Seattle, WA, United States

In this work, we proposed a deep learning structure called SE-UNet for carotid vessel wall segmentation on 3D golden angle radial k-space sampling simultaneous non-contrast angiography and intraplaque hemorrhage (GOAL-SNAP) images. The structure of network consisted of an encoder path for feature extraction and a decoder path for precise localization. The squeeze-and-excitation (SE) module was introduced to the encoder part to learn the context between channels. The proposed SE-UNet achieved high IOU of 0.786, and high pixel-wise sensitivity of 0.976, specificity of 0.850.


Image Reconstruction I

Exhibition Hall
Thursday 13:45 - 14:45
 Acquisition, Reconstruction & Analysis

4744
Computer 26
Fast dynamic speech MRI at 3 Tesla using variable density spirals and constrained reconstruction
Sajan Goud Lingala1, Douglas Blake2, Stanley Kruger3, David Meyer4, Eileen Finnegan2, Ingo Titze2, and Eric Hoffman3

1Department of Biomedical Engineering, University of Iowa, Iowa city, IA, United States, 2Department of Communication Sciences and Disorders, University of Iowa, Iowa city, IA, United States, 3Department of Radiology, University of Iowa, Iowa city, IA, United States, 4Janette Ogg Voice Research Center, Shenandoah University, Winchester, VA, United States

We propose an ultra fast dynamic 3 T MRI scheme for imaging the vocal tract dynamics during speech production. Our approach synergistically exploits efficiency of variable density spirals for motion robustness, artifact suppression, and a sparse SENSE based temporal constrained reconstruction scheme. We realize time resolution of upto 6.2 ms/frame and a spatial resolution of 2.4x2.4 mm2. We demonstrate the utility of this scheme in capturing rapidly varying articulators during fast speech stimuli.

4745
Computer 27
Highly-Accelerated, Real-Time, Phase-Contrast MRI using Radial k-space Sampling and Cartesian GRASP Reconstruction: A Feasibility Study in Pediatric Patients
Hassan Haji-valizadeh1,2, Joshua D. Robinson3,4, Michael Markl2, Cynthia K. Rigsby2,5, and Daniel Kim2

1Department of Biomedical Engineering, Northwestern University, Evanston, IL, United States, 2Department of Radiology, Northwestern University, Chicago, IL, United States, 3Division of Pediatric Cardiology, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States, 4Department of Medical Imaging, Ann & Robert H. Lurie Children's Hospital of Chicago, Chicago, IL, United States, 5Department of Medical Imaging, Northwestern University, Chicago, IL, United States

Iterative compressed sensing reconstruction of real-time phase-contrast MR images acquired with highly-accelerated radial k-space sampling produces considerable image blurring. We propose a Cartesian Golden-angle radial sparse parallel (GRASP) framework that achieves a good balance between image reconstruction speed and data fidelity. The performance of the proposed reconstruction framework is compared with the original GRASP and GROG-GRASP frameworks using 38.4-fold accelerated phase-contrast MRI data acquired from pediatric patients. 

4746
Computer 28
Joint Calibrationless Reconstruction of Highly Undersampled Multi-Contrast MR Datasets Using A Novel Low-Rank Completion Approach
Zheyuan Yi1,2,3, Yilong Liu1,2, Yujiao Zhao1,2, Fei Chen3, and Ed X. Wu1,2

1Laboratory of Biomedical Imaging and Signal Processing, The University of Hong Kong, Hong Kong, China, 2Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong, China, 3Electrical and Electronic Engineering, Southern University of Science and Technology, Shenzhen, China

Routine clinical MRI session often requires multi-contrast imaging with identical geometries but different contrasts, and these images of different contrasts are independently reconstructed despite ubiquitous similarities. Simultaneous autocalibrating and k-space estimation (SAKE) provides a powerful calibrationless parallel imaging approach to reduce scanning time through undersampling. However, traditional SAKE reconstruction does not utilize redundant information embedded in multi-contrast datasets. In this study, we propose to advance SAKE by jointly reconstructing concatenated multi-contrast datasets using a novel low-rank completion approach. Our new method explicitly exploits the correlations in multi-contrast datasets and outperforms the traditional SAKE, leading to higher acceleration factors.  

4747
Computer 29
Non-smooth Convex Optimization for O-Space Reconstruction
Jing Cheng1, Haifeng Wang1, Yuchou Chang2, and Dong Liang1,3

1Lauterbur Research Center for Biomedical Imaging, Shenzhen Institude of Advanced Technoleoy,Chinese Academy of Sciences, Shenzhen, China, 2Department of Computer Science and Engineering Technology, University of Houston - Downtown, Houston, TX, United States, 3Research Center for Medical AI, Shenzhen Institude of Advanced Technoleoy, Chinese Academy of Sciences, Shenzhen, China

Non-linear spatial encoding magnetic (SEM) fields can accelerate data acquisitions and improve the image quality. O-Space imaging generates a radially varying SEM field for spatial encoding in order to achieve more efficient encoding. In this work, we introduce and evaluate a novel primal dual algorithm which can handle the inverse problems of non-smooth convex optimization with non-linear forward operators to reconstruct O-Space images from undersampled data. The experimental results on simulated data show that the proposed method can achieve better image quality compared with the existing methods.

4748
Computer 30
Multi-channel multi-contrast reconstructions via simultaneous use of individual and joint regularization terms
Emre Kopanoglu1,2, Alper Güngör2, Toygan Kilic3,4, Emine Ulku Saritas3,4,5, Kader K. Oguz4,6, Tolga Çukur3,4,5, and H. Emre Güven2

1CUBRIC, School of Psychology, Cardiff University, Cardiff, United Kingdom, 2ASELSAN Research Center, Ankara, Turkey, 3Electrical and Electronics Engineering, Bilkent University, Ankara, Turkey, 4National Magnetic Resonance Research Center (UMRAM), Bilkent University, Ankara, Turkey, 5Neuroscience Program, Sabuncu Brain Research Center, Bilkent University, Ankara, Turkey, 6Radiology, Hacettepe University, Ankara, Turkey

Multi-contrast images of the same anatomy are commonly acquired together to maximize diagnostic information. We demonstrate a multi-channel multi-contrast compressed sensing – parallel imaging (CS-PI) technique that simultaneously uses joint and individual regularization terms to exploit anatomical similarities across contrasts without leakage of distinct features across contrasts and that incorporates coil sensitivities to further improve image quality. The method is compared in-vivo to the single-contrast multi-channel CS-PI method l1-ESPIRiT for PD-/T1-/T2-weighted images of N=11 participants using signal-to-noise ratio calculations as well as neuroradiologist reader studies. The proposed method yields superior performance than l1-ESPIRiT both quantitatively and qualitatively.

4749
Computer 31
Sparse DCE-MRI using a Temporal Constraint Learned from Clinical Data
Sreedevi Gutta1, Yannick Bliesener1, Jay Acharya2, Meng Law2, and Krishna S. Nayak1,2

1Electrical Engineering, University of Southern California, Los Angeles, CA, United States, 2Radiology, University of Southern California, Los Angeles, CA, United States

Dynamic contrast enhanced MRI has benefitted substantially from developments in sparse sampling and constrained reconstruction. Thus far, temporal constraints have proven to be the most powerful. In this work, we explore the use of temporal dictionaries that are learned from a clinical database. We demonstrate that this method provides improved reconstruction quality compared to state-of-the-art TK-model-based constraints or low-rank constraints. The inclusion of spatial information while constructing dictionaries is also explored.

4750
Computer 32
Rapid, Free-Breathing, Cine MRI for Patients with a Cardiac Implantable Electronic Device: A Preliminary Study
KyungPyo Hong1, Jeremy D Collins1,2, Daniel C Lee1,3, and Daniel Kim1,4

1Radiology, Northwestern University Feinberg School of Medicine, Chicago, IL, United States, 2Radiology, Mayo Clinic, Rochester, MN, United States, 3Cardiology, Internal Medicine, Northwestern University Feinberg School of Medicine, Chicago, IL, United States, 4Biomedical Engineering, Northwestern University, Evanston, IL, United States

Standard ECG-gated, breath-held cardiac cine MRI often produces poor image quality in patients with a cardiac implantable electronic device (CIED) due to off-resonance effects, high prevalence of arrhythmia, and/or difficulty in breath-holding. This study seeks to develop a 16-fold accelerated, free-breathing cine MRI pulse sequence using a combination of a gradient echo readout, compressed sensing, and optimal Cartesian k-space sampling. The results from this study shows that an optimal k-space sampling scheme produces superior results compared to random and Poisson disc k-space sampling patterns in imaging phantoms and patients.

4751
Computer 33
Cardiac and Respiratory Motion-Resolved 5D Imaging Using a Free-Running Framework: Comparison of Cartesian and Radial Trajectories
Christopher W Roy1, Jerome Yerly1,2, Jessica AM Bastiaansen1, Nemanja Masala1, Lorenzo Di Sopra1, Jens Wetzl3, Christoph Forman3, Davide Piccini1,4,5, and Matthias Stuber1,2

1Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland, 2Center for Biomedical Imaging (CIBM), Lausanne, Switzerland, 3Siemens Healthcare GmbH, Erlangen, Germany, 4Advanced Clinical Imaging Technology (ACIT), Siemens Healthcare AG, Lausanne, Switzerland, 5LTS5, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland

Recent advances have enabled high resolution cardiac imaging using continuous acquisitions that do not require external gating devices and can be reconstructed in arbitrary dimensions. Here, we extend the use of this Free-running framework to a fully self-gated free-breathing 3D Cartesian trajectory with spiral profile ordering for cardiac and respiratory motion resolved 5D imaging. We demonstrate the feasibility of this Cartesian approach by reconstructing and comparing images from both radial and Cartesian sequences with matching scan parameters in healthy volunteers. Overall, Cartesian images demonstrated comparable cardiac and respiratory motion albeit with more residual artifacts present in the Cartesian images.

4752
Computer 34
Toward single breath-hold whole-heart coverage compressed sensing MRI using VAriable spatial-temporal LAtin hypercube and echo-Sharing (VALAS)
Jingyuan Lyu1, Yu Ding1, Jiali Zhong2, Zhongqi Zhang3, Lele Zhao3, Jian Xu1, Qi Liu1, Ruchen Peng2, and Weiguo Zhang1

1UIH America Inc., Houston, TX, United States, 2Capital Medical University, Beijing LuHe Hospital, Beijing, China, 3Shanghai United Imaging Healthcare Co., Ltd, Shanghai, China

The main goal is to design and implement a sampling and reconstruction strategy that enables full heart coverage in a single breath-hold, with a relatively high spatial resolution (2.5 × 2.5 mm2) and temporal resolution (40 ms). The challenge in sampling pattern design is how to sample most efficiently. In this work, we present a 10 fold accelerated real‐time cardiac cine MRI pulse sequence using a combination of compressed sensing and parallel imaging. 

4753
Computer 35
Non-linear Inverse Compressed-Sensing Reconstruction for Self-Gated Multidimensional Cardiac MRI: XD-NLINV
H. Christian M. Holme1,2, Sebastian Rosenzweig1,2, Xiaoqing Wang1,2, and Martin Uecker1,2

1Institute for Diagnostic and Interventional Radiology, University Medical Center Göttingen, Göttingen, Germany, 2partner site Göttingen, German Center for Cardiovascular Research (DZHK), Göttingen, Germany

Motion is a perpetual challenge in cardiac MRI: for comfortable free-breathing exams, both cardiac and breathing motion need to be resolved. Self-gating approaches have been proposed to automatically bin MRI data into appropriate motion states. Here, we propose a new combined parallel imaging/compressed sensing reconstruction for such multi-dimensional datasets. This method, termed XD-NLINV, solves the non-linear parallel imaging problem, simultaneously estimating images and coil sensitivities. This assures efficient use of the available data and removes the need for pre-calculating the coil profiles. We present initial results showing high image quality for self-gated cardiac short-axis data, resolving both breathing and cardiac motion.

4754
Computer 36
De-Aliasing for Under-sampling in Phase Scrambling Fourier Transform Imaging using Alias-free Reconstruction and Deep Convolutional Neural Network
Satoshi ITO1 and Tsukasa SAITO1

1Utsunomiya University, Utsunomiya, Japan

Alias-free image reconstruction is feasible in phase scrambling Fourier transform imaging. When small down-scaling factor is used in that method, the size of reconstructed images become small and aliased image are separated in the scaled space. In this work, a new fast imaging method in which aliasing artifacts due to under-sampling of signal is removed 2-steps; one is down-scaled space introduced by alias-free reconstruction and the second is the denoising using deep convolution network. It was shown that proposed method provide higher PSNR images compared to random sampling compressed sensing and has an advantage in low sampling rate image acquisition.

4755
Computer 37
Highly Accelerated Simultaneous Multislice Projection Imaging
Nikolai J Mickevicius1, L. Tugan Muftuler2, Andrew S Nencka3, and Eric S Paulson1

1Radiation Oncology, Medical College of Wisconsin, Milwaukee, WI, United States, 2Neurosurgery, Medical College of Wisconsin, Milwaukee, WI, United States, 3Radiology, Medical College of Wisconsin, Milwaukee, WI, United States

Projection imaging has many advantages over Cartesian sampling. The unique point spread function makes it particularly useful for highly accelerated parallel imaging and compressed sensing reconstructions[1]. In this study, a projection-domain sensitivity encoding algorithm is developed for highly accelerated simultaneous multislice radial imaging. Since it operates in the projection-domain, no time expensive gridding, de-gridding, and FFT operations are required within each iteration of the solving algorithm. From an in vivo experiment, two slices were reconstructed from only 34 radial spokes.

4756
Computer 38
Reconstruction of Highly Accelerated Radial Cardiac Cine MRI using GROG based k-t ESPIRiT with TV Constraint
Ibtisam Aslam1, Lindsey A CROWE 2, Miklos KASSAI2, Jean-Paul VALLEE 2, and Hammad Omer1

1Electrical Engineering, COMSATS University Islamabad, Pakistan, Islamabad, Pakistan, 2Service of Radiology, Geneva University Hospitals and Faculty of Medicine, University of Geneva, Switzerland, GENEVA, Switzerland

Breath-hold cardiac cine MRI requires fast data acquisition with good spatio-temporal resolution. Accelerated non-Cartesian trajectories accelerate data acquisition but lead to artifacts. This work proposes a GROG-based k-t ESPIRiT approach with TV to recover the unaliased MR real-time cine images with good spatio-temporal quality. The proposed method was tested on 8 patients with single breath-hold, short-axis, real-time cardiac cine whole-heart stack with under-sampled radially acquired data using trueFISP. The efficiency of the proposed reconstruction was clinically assessed for automated segmentation, CNR & SNR and compared with the standard image reconstruction available on Siemens 3T PRISMA and 1.5T AERA scanners.

4757
Computer 39
Varying Undersampling Dimension for Accelerating Multiple-Acquisition Magnetic Resonance Imaging
Ki Hwan Kim1, Won-Joon Do2, and Sung-Hong Park1

1Bio and Brain Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Korea, Republic of, 2Bio and Brain Engineering, Korea Advanced Institute of Science and Technology, Dajeon, Korea, Republic of

We proposed a new sampling strategy for efficiently accelerating multiple acquisition MRI. The new sampling strategy is to obtain data along different phase encoding directions across multiple acquisitions. The proposed sampling strategy was evaluated in multi-contrast MR imaging (T1, T2, proton density) and multiple phase cycled (PC) balanced steady-state free precession (bSSFP) imaging by using compressed sensing (CS) algorithms and convolutional neural networks (CNNs) with central and/or random sampling pattern. Sampling along different phase encoding directions across multiple acquisitions was advantageous for accelerating multi-acquisition MRI, irrespective of reconstruction method, sampling pattern or datasets, with further improvement through transfer learning.

4758
Computer 40
Sliding Window Reduced FOV Reconstruction in EPI for Real-Time Cardiac Imaging
Patrick Metze1, Tobias Speidel2, Kilian Stumpf1, and Volker Rasche1,2

1Department of Internal Medicine II, University Ulm Medical Center, Ulm, Germany, 2Core Facility Small Animal Imaging (CF-SANI), Ulm University, Ulm, Germany

In this work we present a reconstruction technique based on $$$k$$$-space subtraction of static image parts to acquire real-time cardiac images. The static part is estimated with a sliding window reconstruction of the region outside of the heart to account for respiratory motion. The reduced field of view, i.e. the region of interest, is then reconstructed using a standard SENSE reconstruction, resulting in a temporal resolution of under 40 ms. The image quality is sufficient to estimate functional values in accordance with the BH-CINE reference standard.

4759
Computer 41
High Resolution 3D Isotropic Multi-Contrast Brain Imaging using APIR4EMC
Chaoping Zhang1,2, Alexandra Cristobal-Huerta2, Juan Antonio Hernandez-Tamames2, Stefan Klein1,2, and Dirk H.J. Poot1,2

1Department of Medical Informatics, Erasmus MC, Rotterdam, Netherlands, 2Department of Radiology and Nuclear Medicine, Erasmus MC, Rotterdam, Netherlands

The long scan time of the brain MRI limits its applicability in high resolution 3D isotropic imaging. By using the recent Autocalibrated Parallel Imaging Reconstruction for Extended Multi-Contrast (APIR4EMC) method, we propose a high resolution (1 mm) 3D isotropic multi-contrast (T1, T1-Fatsat, T2, PD, FLAIR) brain imaging method with scan time around 10 min on a 3T MR scanner with an 8-channel brain coil. Experimental results demonstrate the effectiveness of this method.

4760
Computer 42
Simultaneous multislice reconstruction using spiral slice-GRAPPA
Changyu Sun1, Yang Yang2, Daniel S. Weller1,3, Xiaoying Cai1, Craig H. Meyer1, Michael Salerno1,2,4, and Frederick H. Epstein1,4

1Biomedical Engineering, University of Virginia, Charlottesville, VA, United States, 2Medicine, University of Virginia, Charlottesville, VA, United States, 3Electrical and Computer Engineering, University of Virginia, Charlottesville, VA, United States, 4Radiology, University of Virginia, Charlottesville, VA, United States

Spiral trajectories provide efficient data acquisition and favorable motion properties for cardiac MRI. We developed multiband (MB) methods to accelerate spiral cardiac cine imaging including a non-iterative spiral slice-GRAPPA (SSG) reconstruction and a temporal SSG (TSSG). Using 25-35% of k-space for single-band calibration data, experiments in phantoms and five volunteers show 18.7% lower mean artifact power than CG-SENSE when imaging three slices simultaneously. TSSG incorporating CAIPIRINHA with temporal alternation and a temporal filter in reconstruction further reduced rRMSE by 11.2% compared to SSG.     

4761
Computer 43
Accelerated Image Acquisition Using 2D Pulse Segments as Virtual Receivers for GRAPPA
Michael Mullen1,2, Alexander Gutierrez3, Jarvis Haupt4, and Michael Garwood2

1School of Physics and Astronomy, University of Minnesota, Minneapolis, MN, United States, 2Center for Magnetic Resonance Research, Department of Radiology, University of Minnesota, Minneapolis, MN, United States, 3School of Mathematics, University of Minnesota, Minneapolis, MN, United States, 4Department of Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, United States

When based on a k-space description, 2D RF pulses can be applied in segments to increase the excitation bandwidth relative to a single-shot implementation, at a cost of increased imaging time. The increased imaging time can be overcome by undersampling the acquisition in one phase-encoded dimension, where data from each segment are viewed as originating from “virtual receive coils” rather than multiple physical coils. The undersampled data are reconstructed using parallel imaging techniques (e.g. as in GRAPPA). The method was tested in vivo with brain imaging, and the GRAPPA-like reconstruction was comparable in quality to a fully sampled reconstruction.

4762
Computer 44
A Generative Approach to Estimating Coil Sensitivities from Autocalibration data
Yael Balbastre1, Julio Acosta-Cabronero1, Nadège Corbin1, Oliver Josephs1, John Ashburner1, and Martina F Callaghan1

1Wellcome Centre for Human Neuroimaging, UCL Queen Square Institute of Neurology, University College London, London, United Kingdom

We present an algorithm for inferring sensitivities from low-resolution data, acquired either as an external calibration scan or as an autocalibration region integrated into an under-sampled acquisition. The sensitivity profiles of each coil, together with an unmodulated image common to all coils, are defined as penalised-maximum-likelihood parameters of a generative model of the calibration data.  The model incorporates a smoothness constraint for the sensitivities and is efficiently inverted using Gauss-Newton optimization. Using both simulated and acquired data, we demonstrate that this approach can successfully estimate complex coil sensitivities over the full (FOV), and subsequently be used to unfold aliased images.

4763
Computer 45
Efficient MR Image Compression using Deep Learning Models for Multi-contrast MRI
Enhao Gong1,2, Xiaofan Lin3, and John Pauly2

1Subtle Medical, Menlo Park, CA, United States, 2Electrical Engineering, STANFORD UNIVERSITY, Stanford, CA, United States, 3Computer Science and Mathematics, University of California, San Diego, San Diego, CA, United States

As more and more medical imaging dataset is created, efficient and high-rate data compression is in demand for applications such as data transfer, storage and cloud based MR image analysis. However, conventional compression options do not provide the efficiency and compression performance needed for real-time applications such as image query and computer-aided diagnosis. In this work we demonstrated the applicability of the DL based compression algorithm for MRI to improve the compression performance and efficiency. Trained on natural images and fine-tuned on multi-contrast brain MRI, the proposed method provide significantly (~2x) higher compression rate compared with conventional method. Additionally, the end-to-end deep learning compression/de-compression is also several magnitude's faster than conventional methods. This technique can directly benefit industrial and clinical applications, and can provide new model in applications such as multi-contrast fusion and reconstruction.

4764
Computer 46
NAPALM: An Algorithm for MRI Reconstruction with Separate Magnitude and Phase Regularization
Yunsong Liu1 and Justin P. Haldar1

1Ming Hsieh Department of Electrical Engineering, University of Southern California, Los Angeles, CA, United States

We describe a new algorithm for model-based MRI reconstruction with separate magnitude and phase regularization.  The algorithm, named NAPALM, combines the existing proximal alternating linearized minimization (PALM) algorithm for nonsmooth and nonconvex optimization with Nesterov's acceleration and adaptive gradient (AdaGrad) acceleration methods.  Results demonstrate the advantages of NAPALM over existing state-of-the-art algorithms.

4765
Computer 47
Reconstruction Augmentation by Constraining with Intensity Gradients (RACING)
Ali Pour Yazdanpanah1, Onur Afacan1, and Simon K. Warfield1

1Computational Radiology Laboratory, Boston Children's Hospital and Harvard Medical School, Boston, MA, United States

Conventional parallel imaging exploits coil sensitivity profiles to enable image reconstruction from undersampled data acquisition. The extent of undersampling that preserves high quality images is limited in part by the total reduction in signal associated with undersampling, and in part by an additional geometry factor that arises from the position of the coil array with respect to the anatomy being imaged. We propose to further constrain the image reconstruction in order to reduce the geometry factor artifact that limits the use of high acceleration factors.  We derive equality constraints from the acquisition model that induce a coupling between the signal intensity at a voxel, and the signal intensity at other neighboring voxels.  These additional constraints improve the conditioning of the parallel imaging reconstruction by helping to disambiguate aliasing artifact. Consequently, increased undersampling factors have higher image quality. This in turn enables rapid acquisition of 2D and 3D images. Furthermore, these additional constraints are entirely compatible with alternative sources of information to better condition or regularize the image reconstruction. We propose a SENSE formulation of our augmented image reconstruction equations, derive the equality constraints that reduce the g-factor artifact, and solve for the final image using an Augmented Lagrangian numerical formulation.  We also indicate how our formulation can be extended to incorporate image prior models by adding regularized reconstruction or sparsity constraints.

4766
Computer 48
OSCAR-based reconstruction for compressed sensing and parallel MR imaging
Loubna El Gueddari1,2, Emilie Chouzenoux3,4, Jean-Christophe Pesquet4, Alexandre Vignaud1, and Philippe Ciuciu1,2

1CEA/NeuroSpin, Gif-sur-Yvette, France, 2INRIA-CEA Parietal team, Univ. Paris-Saclay, Gif-sur-Yvette, France, 3LIGM, Paris-Est University, Marne-La-Vallée, France, 4CVN, Centrale-Supélec, Univ. Paris-Saclay, Gif-sur-Yvette, France

Compressed sensing combined with parallel imaging has allowed significant reduction in MRI scan time. However, image reconstruction remains challenging and common methods rely on a coil calibration step. In this work, we focus on calibrationless reconstruction methods that promote group sparsity. The latter have allowed theoretical improvements in CS recovery guarantees. Here, we compare the performances of several regularization terms (group-LASSO, sparse group-LASSO and OSCAR) that define with the data consistency term the convex but nonsmooth objective function to be minimized. The same primal-dual algorithm can be used to perform this minimization. Our results demonstrate that OSCAR-based reconstruction is competitive with state-of-the-art $$$\ell_1$$$-ESPIRiT.

4767
Computer 49
An Advanced Optimization Strategy for Joint Estimation of Object and B0
Franz Patzig1, Bertram Wilm1, Kasper Lars1, Maria Engel1, and Klaas Pruessmann1

1University of Zurich and ETH Zurich, Zurich, Switzerland

A major problem of single-shot acquisition techniques are distortions due to local offsets of the static magnetic B0 field. To avoid relying on separately acquired field maps, the object and the B0 map can be jointly estimated, which usually involves updating object and B0 map in an alternating fashion. A new optimization strategy to solve the non-convex B0 sub-problem is suggested. The number of unknowns is significantly reduced by modelling the B0 maps by a smaller basis and a modified version of the simulated annealing algorithm is implemented to better handle the non-convexity. First in-vivo results are presented.

4768
Computer 50
Multi-shot Echo-planar Imaging with Simultaneous MultiSlice Wave-Encoding
JaeJin Cho1, HyunWook Park1, Kawin Setsompop2,3, and Berkin Bilgic2,3

1Korea Advanced institute of Science and Technology, Daejeon, Korea, Republic of, 2Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Boston, MA, United States, 3Harvard Medical School, Boston, MA, United States

We propose an imaging sequence for Simultaneous MultiSlice Multi-Shot EPI (SMS-MS EPI) with wave-CAIPI controlled aliasing, to significantly reduce the imaging time and geometric distortion in gradient echo imaging. We extend the MUSSELS low-rank constrained parallel imaging technique to SMS acceleration and exploit the similarities among the EPI shots for improved reconstruction. In simulations, we demonstrate the capability of our sequence to incorporate wave-CAIPI encoding, which allows higher acceleration rates by fully harnessing the three-dimensional encoding capability of multi-channel receive arrays. Using MUSSELS with wave-SMS, whole-brain T2*-weighted images at 1 mm isotropic resolution can be obtained at the total acceleration of Rtotal=24 (RinplanexRSMS=8x3), corresponding to an acquisition with high image quality and geometric fidelity.


Machine Learning for Image Reconstruction: Optimised

Exhibition Hall
Thursday 13:45 - 14:45
 Acquisition, Reconstruction & Analysis

4769
Computer 51
SANTIS: Sampling-Augmented Neural neTwork with Incoherent Structure for efficient and robust MR image reconstruction
Fang Liu1, Lihua Chen2,3, Richard Kijowski1, and Li Feng4

1Radiology, University of Wisconsin-Madison, Madison, WI, United States, 2Radiology, Southwest Hospital, Chongqing, China, 3Radiology, PLA 101st Hospital, Wuxi, China, 4Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, United States

The purpose of this work was to develop and evaluate a new deep-learning based image reconstruction framework, termed as Sampling-Augmented Neural neTwork with Incoherent Structure (SANTIS) for MR image reconstruction. Our approach combines efficient end-to-end CNN mapping with k-space consistency using the concept of cyclic loss to enforce data fidelity. Adversarial training is implemented for maintaining high quality perceptional image structure and incoherent k-space sampling is used to improve reconstruction accuracy and robustness. The performance of SANTIS was demonstrated for reconstructing vast undersampled Cartesian knee images and golden-angle radial liver images. Our study demonstrated that the proposed SANTIS framework represents a promising approach for efficient and robust MR image reconstruction at vast acceleration rate. 

4770
Computer 52
Crowdsourced Quality Metrics for Image Reconstruction using Machine Learned Ranking
Kevin M Johnson1,2, Laura Eisenmenger3, Patrick Turski2, and Leonardo Rivera-Rivera1

1Medical Physics, University of Wisconsin - Madison, Madison, WI, United States, 2Radiology, University of Wisconsin - Madison, Madison, WI, United States, 3Radiology, University of California - San Francisco, San Francisco, CA, United States

In this work, we investigate a scheme for crowd sourcing image quality using machine learned metrics from user rankings of corrupted images. Using an HTML application, experienced observers ranked pairs of corrupted images with respect to image quality. A convolution neural network (CNN) was then trained to produce a quality score that was higher in the preferred images. The trained CNN was found to be more sensitive to artifacts from image blurring and wavelet compression than mean square error. Finally, preliminary use in training a machine learned image reconstruction is demonstrated.

4771
Computer 53
Virtual Imaging Using Generative Adversarial Networks for Image Translation (VIGANIT): Deep Learning based Prediction of Diffusion-Weighted Images from T2-Weighted Brain MR Images
Vidur Mahajan1, Aravind Upadhyaya2, Vasantha Kumar Venugopal1, Abhishek S Venkataram2, Mukundhan Srinivasan3, Murali Murugavel1, and Harsh Mahajan1,4

1Centre for Advanced Research in Imaging, Neuroscience and Genomics, Mahajan Imaging, New Delhi, India, 2Triocula technologies, Bangalore, India, 3Nvidia, Bangalore, India, 4Mahajan Imaging, New Delhi, India

100 whole brain MRI scans of patients with no abnormality and 30 with acute infarcts, comprising of 25 T2-weighted and Diffusion-Weighted (b=1000) images each, were fed into a Deep Learning model with a 75-25 training-validation split. The T2W image was assigned as the input to predict DW images. Binary Cross entropy of 0.15 for normal and 0.11 for infarct cases was obtained and the predicted images were able to successfully delineate acute and chronic infarcts in all test cases.

4772
Computer 54
A Deep Learning Accelerated MRI Reconstruction Model's Dependence on Training Data Distribution
Dimitrios Karkalousos1, Kai Lønning2, Serge Dumoulin3, Jan-Jakob Sonke4, and Matthan W.A. Caan5

1Spinoza Centre for Neuroimaging, Netherlands, Netherlands, 2Department of Radiation Oncology, the Netherlands Cancer Institute & Spinoza Centre for Neuroimaging, Amsterdam, Netherlands, 3Spinoza Centre for Neuroimaging, Amsterdam, Netherlands, 4Department of Radiation Oncology, the Netherlands Cancer Institute, Amsterdam, Netherlands, 5Academic Medical Center, Amsterdam, Netherlands

Recurrent Inference Machines (RIM) are deep learning inverse problem solvers that have been shown to generalize well to anatomical structures and contrast settings it was not exposed to during training. This makes RIMs ideal for accelerated MRI reconstruction, where the variation in acquisition settings is high. Using T1- and T2*-weighted brain scans and T2-weighted knee scans, we compare the RIM's performance when trained on only a single type of data against the case where all three data types are present in the training set. We present results that show an overall model robustness, but also indicate a slight preference for training on all three types of data.


4773
Computer 55
Exploring the Hallucination Risk of Deep Generative Models in MR Image Recovery
Vineet Edupuganti1, Morteza Mardani1, Joseph Cheng1, Shreyas Vasanawala2, and John Pauly1

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Radiology, Stanford University, Stanford, CA, United States

The hallucination of realistic-looking artifacts is a serious concern when reconstructing highly undersampled MR images. In this study, we train a variational autoencoder-based generative adversarial network (VAE-GAN) on a dataset of knee images and conduct a detailed exploration of the model latent space by generating extensive admissible reconstructions. Our preliminary results indicate that factors such as sampling rate and trajectory as well as loss function affect the risk of hallucinations, but with a reasonable choice of parameters deep learning schemes appear robust in recovering medical images. 

4774
Computer 56
DCTV-Net: Model based Convolutional Neural Network for dynamic MRI
Shanshan Wang1, Yanxia Chen1, Leslie Ying2, Cheng Li1, Ziwen Ke1, Taohui Xiao1, Xin Liu1, Dong Liang1, and Hairong Zheng1

1Shenzhen Institutes of Advanced Technologies, Xili Nanshan, China, 2Department of Biomedical Engineering and Department of Electrical Engineering, The State University of New York, Buffalo, NY, United States

Compressive sensing MRI (CS-MRI) is a popular technique to accelerate MR dynamic imaging. Nevertheless, the reconstruction is normally time-consuming and its parameters have to be hand-tuned To address this challenge, we solve a CS-based dynamic MR imaging problem by adopting the Alternating Direction Method of Multipliers (ADMM) iteration method with the most popular deep learning technique. Specifically, we introduce a deep network structure, dubbed as DCTV-NET, for dynamic magnetic resonance image reconstruction from highly under-sampled k-t space data. Experimental results demonstrate that our method is superior to the state-of-the-art dynamic MRI methods.

4775
Computer 57
Learning Primal Dual Network for Fast MR Imaging
Jing Cheng1, Haifeng Wang1, Leslie Ying2, and Dong Liang1,3

1Lauterbur Research Center for Biomedical Imaging, Shenzhen Institude of Advanced Technoleoy,Chinese Academy of Sciences, Shenzhen, China, 2Department of Biomedical Engineering and Department of Electrical Engineering, University at Buffalo, The State University of New York, Buffalo, NY, United States, 3Research Center for Medical AI, Shenzhen Institude of Advanced Technoleoy, Chinese Academy of Sciences, Shenzhen, China

We introduce a novel deep learning network which combines elements of model and data driven approaches for fast MR imaging, termed modified Learned PD. The network is inspired by the first-order primal dual algorithm, where the convolutional neural network blocks are used to learn the proximal operators. Learned PD network works directly from undersampled k-space data and reconstructs MR images by updating in k-space and image domain alternatively. This approach has been evaluated by in vivo MR datasets and achieves accurate MR reconstructions, outperforming other comparing methods across various quantitative metrics.

4776
Computer 58
Fidelity Imposing Network Edit (FINE) for Solving Ill-Posed Image Reconstruction
Jinwei Zhang1,2, Zhe Liu1,2, Shun Zhang2, Pascal Spincemaille2, Thanh D. Nguyen2, Mert R. Sabuncu1,3, and Yi Wang1,2

1Department of Biomedical Engineering, Cornell University, Ithaca, NY, United States, 2Department of Radiology, Weill Medical College of Cornell University, New York, NY, United States, 3Department of Electrical and Computer Engineering, Cornell University, Ithaca, NY, United States

A Fidelity Imposing Network Edit (FINE) method is proposed for solving inverse problem that edits a pre-trained network's weights with the physical forward model for the test data to overcome the breakdown of deep learning (DL) based image reconstructions when the test data significantly deviates from the training data. FINE is applied to two important inverse problems in neuroimaging: quantitative susceptibility mapping (QSM) and undersampled multi-contrast reconstruction in MRI.

4777
Computer 59
Probabilistic Optimization of Cartesian k-Space Undersampling Patterns for Learning-Based  Reconstruction
Valery Vishnevskiy1, Jonas Walheim1, and Sebastian Kozerke1

1Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland

Learning-based methods offer improved reconstruction accuracy for compressed Sensing MRI. However, most modern methods assume the sampling trajectory to be predefined. In order to further increase reconstruction quality, we present a method for adaptive design of Cartesian undersampling masks. The proposed method delivers sampling trajectories that allow to improve reconstruction accuracy by 26% and 6% compared to the random and state-of-the-art interleaved variable density patterns, respectively.

4778
Computer 60
Deep transform networks for scalable learning of MR reconstruction
Anatole Moreau1,2, Florent Gbelidji1,3, Boris Mailhe1, Simon Arberet1, Xiao Chen1, Marcel Dominik Nickel4, Berthold Kiefer4, and Mariappan Nadar1

1Digital Services, Digital Technology & Innovation, Siemens Medical Solutions, Princeton, NJ, United States, 2EPITA, Le Kremlin-Bicêtre, France, 3CentraleSupélec, Gif-sur-Yvette, France, 4Siemens Healthcare, Application Development, Erlangen, Germany

In this work we introduce RadixNet, a fast, scalable, transform network architecture based on the Cooley-Tukey FFT, and use it in a fully-learnt iterative reconstruction with a residual dense U-Net image regularization. Results show that fast transform networks can be trained at 256x256 dimensions and outperform the FFT.

4779
Computer 61
Automating fetal brain reconstruction using distance regression learning
Lucilio Cordero-Grande1, Anthony N Price1, Emer J Hughes2, Robert Wright3, Mary A Rutherford2, and Joseph V Hajnal1

1Centre for the Developing Brain and Biomedical Engineering Department, School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, 2Centre for the Developing Brain, School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom, 3Biomedical Engineering Department, School of Biomedical Engineering and Imaging Sciences, King's College London, London, United Kingdom

We describe a method for automated fetal brain reconstruction from stacks of 2D single-shot slices. Brain localization is performed by a deep distance regression network. Slice alignment is accomplished by a global search in the rigid transform space followed by registration using a fractional derivative metric. An outlier robust hybrid $$$1,2$$$-norm and linear high order regularization are used for reconstruction. Brain localization has achieved competitive results without requiring annotated segmentations. The method has produced acceptable reconstructions in 129 out of 133 3T fetal examinations tested so far.

4780
Computer 62
AUTOMAP Image Reconstruction of Ultra-Low Field Human Brain MR Data
Koonjoo Neha1,2,3, Bo Zhu1,2,3, Matthew Christensen1,2, John E. Kirsch1,2, and Matthew S Rosen1,2,3

1Department of Radiology, A.A Martinos Biomedical Imaging Center / MGH, Charlestown, MA, United States, 2Harvard Medical School, Boston, MA, United States, 3Department of Physics, Harvard University, Cambridge, MA, United States

Due to very low Boltzmann polarization, MR images acquired at ultra-low field (ULF), MR images require significant signal averaging to overcome low signal-to-noise, which results in longer scan times. Here, we apply the deep neural network image reconstruction technique, AUTOMAP (Automated Transform by Manifold Approximation), to 50% under-sampled low SNR in vivo datasets acquired at 6.5 mT. The performance of AUTOMAP on this data was compared to the conventional 3D Inverse Fast Fourier Transform (IFFT). The results for AUTOMAP reconstruction show a significant improvement in image quality and SNR.  

4781
Computer 63
Synthetic Banding for bSSFP Data Augmentation
Michael Anthony Mendoza1, Nicholas McKibben1, Grayson Tarbox1, and Neal K Bangerter1,2

1Electrical Engineering, Brigham Young University, Provo, UT, United States, 2Bioengineering, Imperial College London, London, United Kingdom

Balanced Steady State Free Precession (bSSFP) MRI is a highly-efficient MRI pulse sequence but suffers from banding artifacts caused by its high sensitivity to magnetic field inhomogeneity. Many algorithms exist that can effectively remove these banding artifacts, typically by requiring multiple phase-cycled acquisitions, which increase scan time. While some of the algorithms can suppress banding to some degree with two sets of phase-cycled acquisitions, much more accurate band suppression is typically achieved with at least four phase-cycled acquisitions. In this work, we present a deep learning method for synthesizing additional phase-cycled images from a set of at least two phase-cycled images that can then be used with existing band reduction techniques in order to reduce scan time.

4782
Computer 64
Magnetic Resonance Fingerprinting Using a Residual Convolutional Neural Network
Pingfan Song1, Yonina C. Eldar2, Gal Mazor2, and Migue Rodrigues1

1Department of EE, University College London, London, United Kingdom, 2Department of EE, Technion, Israel Institute of Technology, Haifa, Israel

Dictionary matching based MR Fingerprinting (MRF) reconstruction approaches suffer from inherent quantization errors, as well as time-consuming parameter mapping operations that map temporal MRF signals to quantitative tissue parameters. To alleviate these issues, we design a residual convolutional neural network to capture the mappings from temporal MRF signals to tissue parameters. The designed network is trained on synthesized MRF data simulated with the Bloch equations and fast imaging with steady state precession (FISP) sequences. After training, our network is able to take a temporal MRF signal as input and directly output corresponding tissue parameters, playing the role of a dictionary and look-up table used in conventional approaches. However, the designed network outperforms conventional approaches in terms of both inference speed and reconstruction accuracy, which has been validated on both synthetic data and phantom data generated from healthy subjects.

4783
Computer 65
A Deep Learning Algorithm for Non-Cartesian Coil Sensitivity Map Estimation
Zihao Chen1,2, Yuhua Chen1,3, Debiao Li1,3, and Anthony G. Christodoulou1

1Biomedical Imaging Research Institute, Cedars-Sinai Medical Center, Los Angeles, CA, United States, 2Department of Engineering Physics, Tsinghua University, Beijing, China, 3Department of Bioengineering, University of California, Los Angeles, Los Angeles, CA, United States

The use of parallel imaging (PI) to exploit the encoding power of multiple coil sensitivity patterns is essential for any modern method for accelerating MRI. In practice, the need to estimate sensitivity maps when using an image-space PI formulation delays the image reconstruction process, particularly for non-Cartesian acquisitions. This paper presents a deep learning method to estimate sensitivity maps from non-Cartesian dynamic imaging data. Results show that this algorithm provide a significant reduction in the time (from 42s to 2.5s for 12 coils) for generating high-quality coil sensitivity maps from non-Cartesian MR data compared to the conventional algorithms.

4784
Computer 66
Real-time MR image reconstruction using Convolutional Neural Networks
Bryson Dietz1, Gino Fallone1,2,3, and Keith Wachowicz1,2

1Oncology, University of Alberta, Edmonton, AB, Canada, 2Medical Physics, Cross Cancer Institute, Edmonton, AB, Canada, 3Physics, University of Alberta, Edmonton, AB, Canada

There has been an increasing interest for systems that combine a linear accelerator with a MRI. The goal of such systems is to allow for real-time adaptive radiotherapy; to have the ability to track a region of interest for the purpose of accurate radiation delivery. This requires the ability to image in real-time. We investigated the use of convolution neural networks (CNNs) for the purpose of real-time imaging. The reconstruction time of our preliminary data was 150 ms using a NVIDIA 1080Ti GTX GPU. Further optimization of the CNN parameters may decrease the reconstruction time below 100 ms.

4785
Computer 67
ShiftNets: Deep Convolutional Neural Networks for MR Image Reconstruction & the Importance of Receptive Field of View
Philip K. Lee1,2, Makai Mann1, and Brian A. Hargreaves1,2,3

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Radiology, Stanford University, Stanford, CA, United States, 3Bioengineering, Stanford University, Stanford, CA, United States

Deep learning has been applied to the Parallel Imaging problem of resolving coherent aliasing in image domain. Convolutional neural networks have finite receptive FOV, where each output pixel is a function of a limited number of input pixels. For uniformly undersampled data, a simple hypothesis is that including the aliased peak in the receptive FOV would improve suppression of aliasing. We show that a simple channel augmentation scheme allows us to resolve aliasing using 50x fewer parameters than a large U-Net with millions of parameters and a global receptive FOV. This method was tested on retrospectively undersampled knee volumes.

4786
Computer 68
POCS Augmented CycleGAN for MR Image Reconstruction
Hanlu Yang1, Yiran Li1, Danfeng Xie1, and Wang Ze2

1Electrical & Computer Engineering Department, Temple University, Philadelphia, PA, United States, 2Department of Radiology, Temple University, Philadelphia, PA, United States

Traditional MRI reconstruction depends heavily on solving nonlinear optimization problems, which could be highly time-consuming and sensitive to noise. We proposed a hybrid DL-based MR image reconstruction method by combining two state-of-art deep learning networks, U-Net and CycleGAN (Generative adversarial network with cycle loss) and a traditional method: projection onto convex set (POCS). Our result shows a high reconstruction accuracy and this method can be further used to increase the sample size, which may find many applications in situations where the training samples are limited such as medical images.  

4787
Computer 69
Accelerated Targeted Coronary MRI Using Sparsity-Regularized SPIRiT-RAKI
Seyed Amir Hossein Hosseini1,2, Steen Moeller2, Sebastian Weingärtner1,2,3, Kâmil Uğurbil2, and Mehmet Akçakaya1,2

1Electrical and Computer Engineering, University of Minnesota, Minneapolis, MN, United States, 2Center for Magnetic Resonance Research (CMRR), University of Minnesota, Minneapolis, MN, United States, 3Computer Assisted Clinical Medicine, University Medical Center Mannheim, Heidelberg University, Mannheim, Germany

Long scan duration remains a challenge in coronary MRI. A scan-specific machine learning technique, called Robust Artificial-neural-network for k-space Interpolation (RAKI) has recently shown promising results in accelerating MRI. However, RAKI was originally designed for uniform undersampling patterns. In this study, we propose a technique, called SPIRiT-RAKI that enables RAKI with arbitrary undersampling using scan-specific convolutional neural networks to enforce self-consistency among coils. Regularization terms are also incorporated in the new formulation. Our results indicate that SPIRiT-RAKI can successfully accelerate 3D targeted coronary MRI. 

4788
Computer 70
A divide-and-conquer strategy to overcome memory limitations of current GPUs for high resolution MRI reconstruction via a domain transform deep learning method
Chengzhu Zhang1, Dalton Griner1, Yinsheng Li1, Yijing Wu1, and Guang-hong Chen1,2

1Medical Physics, University of Wisconsin-madison, Madison, WI, United States, 2Radiology, University of Wisconsin-madison, Madison, WI, United States

Direct learning of a domain transform to reconstruct images with flexible data acquisition schemes represents a step to achieve intelligence in image reconstruction. However, a technical challenge that is encountered with the domain transform type of learning strategy is that current network architectures and training strategies are GPU memory hungry. As a result, given the currently available GPUs with memory on the order of 24 GB, it is very difficult to achieve high resolution (beyond 128x128) MRI reconstruction. The main purpose of this paper is to present a divide-and-conquer strategy to reconstruct high resolution (better than 256x256) MRI images via domain transform learning while staying within the current GPU memory restrictions.

4789
Computer 71
A New Deep Learning Structure for Improving Image Quality of a Low-field Portable MRI System
WENCHUAN MU1, Liang Zheng2, Danial C. Alexander3, Jia Gong1, Wenwei Yu4, and Shao Ying Huang1,5

1Engineering Product Development, Singapore University of Technology and Design, SINGAPORE, Singapore, 2Information Systems Technology and Design, Singapore University of Technology and Design, SINGAPORE, Singapore, 3Centre for Medical Image Computing and Dept. Computer Science, UCL, London, United Kingdom, 4Center for Frontier Medical Engineering, Chiba University, Chiba, Japan, 5Department of Surgery, National University of Singapore, Singapore, Singapore

A permanent magnet based low-field MRI system provides portability and affordability. However, the quality of the image is low due to a low signal-to-noise ratio (SNR). We propose a new deep learning structure which effectively integrates denoising-networks end-to-end to super-resolution-networks, to transfer the rich information available from one-off experimental imaging from a mid-field MRI scanner (1.5T) to the lower-quality data from a portable system. The procedure uses matched pairs to learn mappings from low-quality to the corresponding high-quality images. Using the proposed method, the quality and resolution of an image from a low-field MRI system is significantly improved.

4790
Computer 72
Simultaneous Multi-Slice Deep RecOnstruction NEtwork (SMS-DRONE)
Ouri Cohen1

1Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY, United States

Recently, MR fingerprinting (MRF) has been proposed as a means of disentangling simultaneously excited slices by exciting each slice with a distinct acquisition schedule. A notable drawback of this approach, which is particularly acute for multi-parametric dictionaries, is the linear increase in reconstruction time with the number of slices and the potential reduction in accuracy. Here we describe an extension to our previously described MRF-DRONE method that can overcome these issues. Our method can enable larger acceleration factors and faster reconstruction of multi-parametric data.

4791
Computer 73
Deep Learning Super-FOV for Accelerated bSSFP Banding Reduction
Nicholas McKibben1, Michael Mendoza1, Edward DiBella2, and Neal K. Bangerter3

1Electrical and Computer Engineering, Brigham Young University, Provo, UT, United States, 2Radiology, University of Utah, Salt Lake City, UT, United States, 3Imperial College London, London, United Kingdom

We present a technique for bSSFP band removal using two undersampled phase-cycled bSSFP image acquisitions.

4792
Computer 74
Convolutional Neural Network for Real-Time High Spatial Resolution Functional Magnetic Resonance Imaging
Cagan Alkan1, Zhongnan Fang2, and Jin Hyung Lee1,3,4,5

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2LVIS Corporation, Palo Alto, CA, United States, 3Neurology, Stanford University, Stanford, CA, United States, 4Bioengineering, Stanford University, Stanford, CA, United States, 5Neurosurgery, Stanford University, Stanford, CA, United States

We propose a convolutional neural network (CNN) based real-time high spatial resolution fMRI method that can reconstruct a 3D volumetric image (140x140x28 matrix size) in 150 ms. We achieved 4x spatial resolution improvement using variable density spiral (VDS) trajectory design. The proposed method achieves similar reconstruction performance as our earlier compressed sensing reconstructions while achieving 17x faster reconstruction time. We demonstrate that this method accurately detects cortical layer specific activity.

4793
Computer 75
Spatio-Temporal Undersampling Artefact Reduction with Neural Networks for Fast 2D Cine MRI with Limited Data
Andreas Kofler1, Marc Dewey1, Tobias Schaeffter2,3, Christian Wald1, and Christoph Kolbitsch2,3

1Department of Radiology, Charité - Universitätsmedizin Berlin, Berlin, Germany, 2Physikalisch-Technische Bundesanstalt (PTB), Braunschweig and Berlin, Germany, 3School of Biomedical Engineering and Imaging Sciences, King’s College London, London, United Kingdom

A well-known bottleneck of neural networks is the requirement of large datasets for successful training. We present a method for reduction of 2D radial cine MRI images which allows to properly train a neural network on limited datasets. The network is trained on spatio-temporal slices of healthy volunteers which are previously extracted from the image sequences and is tested on patients data with known heart dysfunction. The image sequences are reassembled from the processed spatio-temporal slices. Our method is shown to have several advantages compared to other Deep Learning-based methods and achieves comparable results to a state-of-the-art Compressed Sensing-based method.


Segmentation 2

Exhibition Hall
Thursday 13:45 - 14:45
 Acquisition, Reconstruction & Analysis

4794
Computer 76
Automatic Segmentation Of The Myocardium in Cardiac Arterial Spin Labelling Images Using a Deep Learning Model Facilitates Myocardial Blood Flow Quantification
Pedro M. Gordaliza1,2, Verónica Aramendía‐Vidaurreta3, Juan José Vaquero1,2, Gorka Bastarrika3, María Asunción Fernández-Seara3, and María Arrate Muñoz-Barrutia1,2

1Bioengineering, Universidad Carlos III de Madrid, Leganés, Spain, 2Instituto de Investigación Sanitaria Gregorio Marañón, Madrid, Spain, 3Radiology Deparment, Clínica Universidad de Navarra, Pamplona, Spain

Arterial Spin Labelling (ASL) allows to quantify Myocardial Blood Flow (MBF) by averaging over multiple ASL pairs. However, the procedure heavily depends on the manual segmentation of the myocardium. In this work, we introduce a Deep Learning model to segment this region and build a completely automatic pipeline for the  MBF estimation. The accomplished evaluation results prove the success of the proposed method, which presents: 1) high overlap between the automatically extracted masks and those manually segmented by an expert (Dice Similarity Coefficient around 90%) and 2) good agreement of the MBF estimations with those obtained from the manual annotations.

4795
Computer 77
Spinal Cord Grey Matter Segmentation using a Light-Weight Off-The-Shelf Neural Network
Jackie Yik1,2, Roger Tam3,4, John K. Kramer2,5, Cornelia Laule1,2,4,6, and Hanwen Liu1,2

1Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada, 2International Collaboration on Repair Discoveries, Vancouver, BC, Canada, 3School of Biomedical Engineering, University of British Columbia, Vancouver, BC, Canada, 4Radiology, University of British Columbia, Vancouver, BC, Canada, 5Kinesiology, University of British Columbia, Vancouver, BC, Canada, 6Pathology and Laboratory Medicine, University of British Columbia, Vancouver, BC, Canada

Spinal cord grey matter segmentation is typically done manually. Automatic segmentation methods exist but are generally highly customized. We used an off-the-shelf neural network (LinkNet) to segment the grey matter in the spinal cord to assess the performance of a method with a generic architecture, which may be easier to replicate on different machine learning frameworks. Manual segmentation was used as training data. The performance of our trained network was compared to an automatic segmentation method in the Spinal Cord Toolbox (SCT), and both networks produced similar results, demonstrating the viability of the off-the-shelf approach.

4796
Computer 78
Feasibility of brain white matter segmentation on multi-echo T2-weighted images without registration: a Neural Network approach.
Jackie Yik1,2, Roger Tam3,4, Cristina Rubino5, Lara Boyd6, David K.B. Li4,7, Cornelia Laule1,2,4,8, and Hanwen Liu1,2

1Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada, 2International Collaboration on Repair Discoveries, Vancouver, BC, Canada, 3School of Biomedical Engineering, University of British Columbia, Vancouver, BC, Canada, 4Radiology, University of British Columbia, Vancouver, BC, Canada, 5Rehabilitation Sciences, University of British Columbia, Vancouver, BC, Canada, 6Physical Therapy, University of British Columbia, Vancouver, BC, Canada, 7Medicine, University of British Columbia, Vancouver, BC, Canada, 8Pathology and Laboratory Medicine, University of British Columbia, Vancouver, BC, Canada

Most current methods of human brain white matter segmentation require registration to Timage space. Artificial intelligence can reduce potential errors in, and speed up, this process by segmenting white matter from T2-weighted images directly.  A neural network was pre-trained using T1-weighted images and FSL’s FAST followed by T2-weighted images using transfer learning. The network could then segment new T2-weighted images directly. T1- and T2-weighted image segmentations using the neural network were comparable to FSL’s FAST. Our work shows the feasibility of multi-echo T2-weighted images for brain white matter segmentation without initial segmentation and registration of T1-weighted images.

4797
Computer 79
Automated Fetal Brain Segmentation Using Deep Convolutional Neural Network
Bin Chen1, Liming Wu1, Bing Zhang2, Simin Liu3, and Hua Guo3

1Purdue University Northwest, Hammond, IN, United States, 2Nanjing University Medical School, Nanjing, China, 3Tsinghua University, Beijing, China

Recent advances show promising fetal brain reconstruction results through image motion correction and super resolution from a stack of unregistered images consisting of in-plane motion free snapshot slices acquired by fast imaging methods. Most motion correction and super resolution techniques for 3D volume reconstruction require accurate fetal brain segmentation as the first step of image analysis. In this study, a customized U-Net based deep learning method was implemented for automatic fetal brain segmentation. The high accuracy of deep learning based semantic segmentation improves the performance in volume registration as well as quantitative studies of brain development and group analysis.

4798
Computer 80
Random forests and DenseNet: a comparative study of brain gliomas segmentation
Marco Castellaro1,2, Gianmario Battista2, and Alessandra Bertoldo1,2

1Padova Neuroscience Center, University of Padova, Padova, Italy, 2Department of Information Engineering, University of Padova, Padova, Italy

Machine Learning techniques can provide useful automatic tools. Segmentation of brain tumors is a time consuming task that could potentially beneficiate from its automation. This work investigate and compare the performances of two frameworks: Random forest and DenseNet. The former is a well known framework and the latter is a novel technique based on deep learning.

4799
Computer 81
Hypothalamus semi-automatic segmentation from MR images using Convolutional Neural Networks
Lívia Rodrigues1, Thiago Rezende2, Ariane Zanesco2, Ana Luiza Hernandez2, Marcondes França2, and Letícia Rittner1

1Medical Image Computing Lab, School of Electrical and Computer Engineering (FEEC), University of Campinas, Campinas, Brazil, 2Department of Neurology, School of Medical Sciences, University of Campinas, Campinas, Brazil

Hypothalamus is a small structure of the brain with important role in sleep, body temperature regulation and emotion. Some diseases as schizophrenia can be attributed to volumetric change on hypothalamus, usually measured through Magnetic Resonance Imaging (MRI). However, hypothalamic morphological landmarks are not always clear and manual segmentation can become variable, leading to inconsistent data on literature. On this project, hypothalamus was automatically segmented using convolutional neural networks (CNNs) . Three independent CNNs were trained, one for each view of volumetric MRI, obtaining final dice of 0.787 for axial view, 0.781 for sagittal and 0.747 for coronal view. 

4800
Computer 82
Automatic Segmentation of Brain Metastases Using Saturation Transfer Magnetic Resonance Imaging
Elham Karami1,2,3, Wilfred Lam2, Wendy Oakden2, Margaret Koletar2, Leedan Murray2, Stanley Liu1,4,5,6, Ali Sadeghi Naini1,2,3,6, Hany Soliman1,6, Arjun Sahgal1,6, and Greg Stanisz1,2,7

1Medical Biophysics, University of Toronto, Toronto, ON, Canada, 2Physical Sciences Platform, Sunnybrook Research Institute, Toronto, ON, Canada, 3Electrical and Computer Engineering, Lassonde School of Engineering, York University, Toronto, ON, Canada, 4Radiation Oncology, University of Toronto, Toronto, ON, Canada, 5Biological Sciences Platform, Sunnybrook Research Institute, Toronto, ON, Canada, 6Radiation Oncology, Sunnybrook Health Sciences Centre, Toronto, ON, Canada, 7Neurosurgery and Paediatric Neurosurgery, Medical University of Lublin, Lublin, Poland

Chemical exchange saturation transfer (CEST) and magnetization transfer (MT) are MR contrast mechanisms that have been shown to correlate with cancer metabolism. Given that CEST does not require exogenous contrast agents, the goal of this study was to investigate the potential of CEST for segmenting the images of brain metastasis. As such, the tumour, and edema were segmented on CEST images and compared with segmentation performed on FLAIR and post-gadolinium T1-weighted images. The results indicate that the Dice similarity coefficient ranges between 0.78 to 0.84, suggesting that CEST can potentially be used for segmentation of brain metastases.

4801
Computer 83
Segmentation of Intra-Tumour Distinct Metabolic Regions Using Chemical Exchange Saturation Transfer Imaging
Elham Karami1,2,3, Wilfred Lam2, Wendy Oakden2, Margaret Koletar2, Leedan Murray2, Stanley Liu1,4,5,6, Ali Sadeghi Naini1,2,3,6, and Greg Stanisz1,2,7

1Medical Biophysics, University of Toronto, Toronto, ON, Canada, 2Physical Sciences Platform, Sunnybrook Research Institute, Toronto, ON, Canada, 3Electrical and Computer Engineering, Lassonde School of Engineering, York University, Toronto, ON, Canada, 4Radiation Oncology, University of Toronto, Toronto, ON, Canada, 5Biological Sciences Platform, Sunnybrook Research Institute, Toronto, ON, Canada, 6Radiation Oncology, Sunnybrook Health Sciences Centre, Toronto, ON, Canada, 7Neurosurgery and Paediatric Neurosurgery, Medical University of Lublin, Lublin, Poland

Chemical exchange saturation transfer (CEST) is a promising MR contrast mechanism that has been shown to correlate with cancer metabolism and reveal regions of active tumour metabolism. However, the acquisition of CEST-weighted images is time consuming. In this study, computational methods including unsupervised learning were adapted to find the minimum number of CEST images required to segment the intra-tumour distinct metabolic regions accurately, and to find the number of different cell groups existing within a tumour. The results indicate that four intra-tumour regions can be segmented accurately using only CEST images acquired at 3.5 ppm and 2.0 ppm.

4802
Computer 84
Automatic Glioma Segmentation Algorithm Based on Superpixel Features
Yaping Wu1,2, Yusong Lin2, Guohua Zhao2, Longfei Li2, and Meiyun Wang3

1School of Electronic and Information Engineering, Xi’an Jiaotong University, Xi'an, China, 2Collaborative Innovation Center for Internet Healthcare and School of Software and Applied Technology, Zhengzhou University, Zhengzhou, China, 3Department of Radiology, Henan Provincial People’s Hospital, Zhengzhou, China

This study proposes an algorithm to locate and segment Glioma tumor automatically. The algorithm contains three main steps. Firstly, a self-adaptation simple linear iterative clustering (ASLIC0) algorithm was executed to segment T2 weighted MRI images to superpixels images. Then, 52 features including fractal features, curvature feature and higher order derivative map Haralick texture features was calculated on each superpixel. Finally, a Support Vector Machine was trained as a classifier to select superpixels belong to tumor lesion or not. The Dice overlap measure for the segmented Glioma is 0.87 on the data set from the Henan Provincial People’s Hospital. 

4803
Computer 85
Machine Learning-based Human Knee Cartilage Segmentation on MRI
Siddhi Munde1, Melissa N Manzer1, Wellsandt Elizabeth2, Jessica Emory3, and Balasrinivasa R Sajja1

1Radiology, University of Nebraska Medical Center, Omaha, NE, United States, 2Division of Physical Therapy, University of Nebraska Medical Center, Omaha, NE, United States, 3University of Nebraska Medical Center, Omaha, NE, United States

Accurate knee cartilage segmentation on MRI is essential to obtain quantitative measures from cartilage that help in the assessment of knee pathology and therapeutic response in patients with diseases such as Osteoarthritis. Segmentation of cartilage on routine clinical MRI is challenging due to image intensity variation across the structure and low image contrast. In this study, we obtained an accurate cartilage segmentation on PD and T1 weighted images using Support Vector Machine (SVM) classifier with a spatial indexing  feature which accounts for regional signal variations.

4804
Computer 86
Automated Segmentation of Thalamic Nuclei using Convolutional Neural Networks
Mohammad Sadegh Majdi1, Mahesh Bharath Keerthivasan2, Natalie M Zahr3, Jeffrey J Rodriguez1, and Manoj Saranathan4

1electrical and computer engineering, university of arizona, tucson, AZ, United States, 2university of arizona, tucson, AZ, United States, 3stanford, stanford, CA, United States, 4medical imaging, university of arizona, tucson, AZ, United States

parcellation of thalamic nuclei is critical step in  targeting for deep brain surgery, volumetry for longitudinal tracking of diseases such as  Alzheimer’s and multiple sclerosis,. However, thalamic nuclei are mostly indistinguishable in conventional T1 or T2 weighted MRI. In this study, we propose a deep neural network based method to achieve a fast and accurate segmentation of thalamic nuclei, taking advantage of high contrast characteristics of a white matter nulled MPRAGE sequence at both 3T and 7T.


4805
Computer 87
Machine Learning Techniques for Bone Tumor Segmentation using Diffusion MRI
Sneha Patil1, Esha Baidya Kayal1, Sameer Bakhshi2, Raju Sharma3, Devasenathipathy Kandasamy3, and Amit Mehndiratta1,4

1Center for Biomedical Engineering, Indian Institute of Technology, Delhi, India, New Delhi, India, 2Medical Oncology, IRCH, All India Institute of Medical Sciences, New Delhi, India, New Delhi, India, 3Radio Diagnosis, All India Institute of Medical Sciences, New Delhi, India, New Delhi, India, 4Department of Biomedical Engineering, All India Institute of Medical Sciences, New Delhi, India, New Delhi, India

Automatic and accurate segmentation of osteosarcoma region in MRI images can assist doctor to prepare a feasible treatment plan, hence resulting in improved cure rate. The purpose of this study was to evaluate and compare the performance of automated and semi-automated algorithms that might be effective in segmenting bone tumor in MRI data, with reasonable accuracy, speed and minimal manual input. The results are very conclusive for efficient performance.

4806
Computer 88
Can Intensity Augmentation Improve Generalizability of CNN-based Image Segmentation?
Nina Jacobsen1, Andreas Deistung1,2, Dagmar Timmann2, Sophia Luise Goericke3, Jürgen R. Reichenbach1,4, and Daniel Güllmar1

1Medical Physics Group, Institute for Diagnostic and Interventional Radiology, Jena University Hospital, Jena, Germany, 2Department of Neurology, Essen University Hospital, University of Duisburg-Essen, Essen, Germany, 3Department of Diagnostic and Interventional Radiology and Neuroradiology, University of Duisburg-Essen, Essen, Germany, 4Michael-Stifel-Center-Jena for Data-Driven and Simulation Science, Friedrich-Schiller University Jena, Jena, Germany

As a strategy to achieve higher generalizability of a convolutional neural network (CNN), data augmentation may be used to introduce a higher degree of variability within the training sample. In this study, five different intensity augmentation strategies were compared and analyzed by means of the CNN segmentation performance. The results indicate how intensity augmentation improves the robustness, and thereby the generalizability, of the CNN but in some cases also compromise the segmentation performance in terms of accuracy.

4807
Computer 89
U-Net Segmentation for Human Body Models for SAR Simulations
Isabelle Heukensfeldt Jansen1, Matthew Tarasek1, Johan Reimann1, and Desmond Teck Beng Yeo1

1General Electric, Niskayuna, NY, United States

RF power absorption during MRI, expressed in terms of specific absorption rate (SAR), is an important safety issue, especially in multi-channel transmit MRI. To reduce uncertainties of local SAR estimates due to subject antatomical variations, patient-specific human body models can be applied in EM simulations of the RF transmit coil. In this work, we trained a U-net neural network on simulated CT scans to quickly create HBMs with four primary tissue classes (bone, lungs, fat, and water-based). Local SAR results using HBMs created with the U-net showed good agreement with those from ground truth models.

4808
Computer 90
Automated Knee MRI Semantic Segmentation with Generative Adversarial Networks
Dimitri A Kessler1, James W MacKay1, Martin J Graves1, Fiona J Gilbert1, and Joshua D Kaggie1

1Department of Radiology, University of Cambridge, Cambridge, United Kingdom

We describe a fully automated deep learning approach for generating semantic segmentation maps of the knee joint. A conditional Generative Adversarial Network (cGAN) was trained on 3D fat-saturated spoiled gradient recalled-echo MRIs of the knee from nine individuals (nimages=778) to generate segmentation maps containing the patella, femur and tibia. The trained network was tested with a separate dataset of one individual (nimages=80). The mean Sørensen–Dice Similarity Coefficient (DSC) was 0.959 and Jaccard Index was 0.985 for all three compartments. These results suggest that cGANs can perform accurate bony segmentation of the knee.

4809
Computer 91
Volumetric Segmentation of Acute Brain Infarcts on Diffusion-Weighted Imaging using Deep Learning
Ken Chang1, James Brown1, Andrew L Beers1, Katharina Hoebel1, Jay Patel1, Otto Rapalino2, Bruce Rosen1, Hakan Ay1, and Jayashree Kalpathy-Cramer2

1Radiology, Massachusetts General Hospital, Boston, MA, United States, 2Massachusetts General Hospital, Boston, MA, United States

Rapid and accurate evaluation of stroke is imperative as currently available treatments are constrained by a narrow time window. Diffusion Weighted Magnetic Resonance (DWI) is a key imaging modality in stroke evaluation as it allows for assessment of the extent of acute ischemic brain injury.  Nonetheless, manual delineation of stroke regions is expensive, time-consuming, and subject to inter-rater variability.  In this study, we sought to develop a deep learning approach for ischemic stroke volumetric segmentation in a large clinical dataset of 1,205 patients from the NIH-funded Heart-Brain Interactions in Human Acute Ischemic Stroke Study utilizing only DWI imaging.

4810
Computer 92
Real-time Ultrafast Fetal Brain Localization using Convolutional Neural Networks
Dhineshvikram Krishnamurthy1, Wonsang You1, Kushal Kapse1, and Catherine Limperopoulos1

1Division of Diagnostic Imaging and Radiology, Children's National Health System, Washington, DC, United States

The advent of fetal magnetic resonance imaging has provided innovative approaches to study in-vivo brain development in the womb. One of the major challenges in the quantification of fetal brain growth and development is regional and tissue-specific segmentation. An automated brain localization algorithm can reduce the time and facilitate accurate segmentation. In this study, we propose an ultra-fast and robust method for fetal brain localization from SSFSE anatomical images using a minimally modified object localization algorithm called You Only Look Once (YOLO). YOLO provides not only the enhanced accuracy of brain localization by differentiating brain from maternal tissues but also fast computation time for brain detection compared to the other algorithms.

4811
Computer 93
Target-class-agnostic feature rejection for radiomics analyses based on variations of tumor segmentation mask
Balthasar Schachtner1, Michael Ingrisch1, Gresser Eva1, Moritz Schneider1, Andrea Schreier1, Olga Solyanik1, Guiseppe Magistro2, and Dominik Nörenberg1

1Department of Radiology, Munich University Hospitals, LMU, Munich, Germany, 2Department of Urology, Munich University Hospitals, LMU, Munich, Germany

Feature selection is a key aspect to radiomics analyses. An approach to remove features which are not stable with respect to small variations of the segmented mask is presented. The rejection works target-class agnostic and can be used in combination with target-class-based selections. An increase of about 5 percentage points can be seen when using the proposed approach in a simple machine learning setup on prostate MRI of prostate cancer patients.

4812
Computer 94
Carotid Artery Localization and Lesion Classification on 3D-MERGE MRI using Neural Network and Object Tracking methods
Li Chen1, Huilin Zhao1,2, Niranjan Balu1, Xihai Zhao3, Rui Li3, Jianrong Xu2, Thomas S Hatsukami1, Jenq-Neng Hwang1, and Chun Yuan1

1University of Washington, Seattle, WA, United States, 2Radiology, Renji Hospital, School of Medicine, Shanghai Jiao Tong University, Shanghai, China, 3Department of Biomedical Engineering, Tsinghua University School of Medicine, Beijing, China

Carotid vessel wall imaging (VWI) with MRI provides additional prognostic value for cerebro/cardiovascular ischemic events, beyond current clinical diagnostic imaging methods. While fast 3D carotid MRI is possible, manual review of the large 3D dataset is time consuming. Automatic identification of artery locations and lesion categories are therefore required for VWI screening protocols. With neural network and object tracking methods, we developed a fully automated analysis tool to find common/internal/external carotid arteries and flag possible high-risk lesion locations. The tool achieved 0.782 Intersection over Union (IoU) for artery localization, and 0.895 sensitivity for high-risk lesion classification.

4813
Computer 95
Relevance-guided Feature Extraction for Alzheimer's Disease Classification
Christian Tinauer1, Stefan Heber1, Lukas Pirpamer1, Anna Damulina1, Maximilian Sackl1, Edith Hofer1, Marisa Koini1, Reinhold Schmidt1, Stefan Ropele1, and Christian Langkammer1

1Department of Neurology, Medical University of Graz, Graz, Austria

Using FLAIR images we separated Alzheimer's patients (n=106) from controls (n=173) by using a deep convolutional neural network and found that the classifier might learn irrelevant features e.g. outside the brain. Preprocessing of MRI plays a crucial but often neglected role in classification and therefore we have developed a method enforcing the relevant features to be within brain tissue and, thus, eliminated the influence of precomputed brain masks. While our relevance-guided training method reached the same classification accuracy, incorporating relevance improved feature identification in an anatomically more reasonable manner.

4814
Computer 96
Classification of benign and malignant lymph nodes based on ex-vivo diffusion MRI data
Andrada Ianus1,2, Inês Santiago2, Daniele Ravi1, Celso Matos2, Daniel C. Alexander1, and Noam Shemesh2

1Centre for Medical Image Computing, Department of Computer Science, University College London, London, United Kingdom, 2Champalimaud Research, Champalimaud Centre for the Unknown, Lisbon, Portugal

Developing non-invasive imaging technique for detection and characterisation of lymph nodes is an important topic in cancer research. Diffusion MRI (dMRI) appears to be a promising modality for this task. This work investigates the ability of dMRI to differentiate benign and malignant lymph nodes based on a rich, ex-vivo dataset, and aims to find which measurements provide the most differentiation power.  

4815
Computer 97
Automatic classification of benign and malignant prostate lesions: A comparison using VERDICT DW-MRI and ADC maps
Eleni Chiou1,2, Edward Johnston3, Francesco Giganti4,5, Elisenda Bonet-Carne1, Shonit Punwani3, Iasonas Kokkinos2, and Eleftheria Panagiotaki1,2

1UCL Centre for Medical Image Computing, Department of Computer Science, University College London, London, United Kingdom, 2Department of Computer Science, University College London, London, United Kingdom, 3UCL Centre for Medical Imaging, Division of Medicine, University College London, London, United Kingdom, 4Department of Radiology, UCLH NHS Foundation Trust, University College London, London, United Kingdom, 5Division of Surgery & Interventional Science, University College London, London, United Kingdom

Currently, many studies exploit deep learning and mp-MRI data to enhance the diagnostic accuracy of prostate cancer characterisation. In this study, we focus on VERDICT DW-MRI data and compare its diagnostic performance to those of the ADC map and the raw DW-MRI from the mp-MRI.  Specifically, we compare the performance obtained by a fully convolutional neural network (CNN) when training and test is performed on the raw VERDICT DW-MRI, the ADC maps and the DW-MRI data from the mp-MRI acquisition. The results indicate that the CNN performs better when it is trained and tested on VERDICT DW-MRI.


4816
Computer 98
MRI texture analysis for detection of axillary lymph node metastasis in breast cancer patients
Renee Cattell1, Vincent Zhang1, Pauline Huang1, Meghan Italo1, James Kang1, Jason Ha1, Haifang Li1, Jules Cohen1, Lea Baer1, Dinko Franceschi1, Cliff Bernstein1, Sean Clouston1, and Timothy Duong1

1Stony Brook University, Stony Brook, NY, United States

We tested the hypothesis whether texture analysis of axillary lymph node (aLN) MRI can reliably detect cancer metastasis in the aLN. Comparison was made with ground truth based on pathology and clinical reports. The top single-feature predictor yielded an area under the curve (AUC) of 0.91 and the top two-feature combination yielded an AUC of 0.95. These findings showed that texture analysis of aLN MRI can accurately predict disease status in the nodes associated with breast cancer. 

4817
Computer 99
Robust detection of anatomical landmark by combining adaptive boosting and active shape model for automated scan plane planning of spine MRI
Suguru Yokosawa1, Yoshimi Noguchi1, Kenta Sakuragi2, Kuniaki Harada2, Masatomo Yokose2, Hisako Nagao2, and Hisaaki Ochi1

1Research & Development Group, Hitachi, Ltd., Tokyo, Japan, 2Healthcare Business Unit, Hitachi, Ltd., Tokyo, Japan

 Automated scan plane planning is expected to improve MRI scanner usability and provide consistent scan plane prescriptions which are useful for follow-up examinations. However, a landmark degenerated by formation of a lesion such as an intervertebral disc in the case of hernia patient is difficult to detect because shape and properties of tissue greatly deviate from normal cases. In this study, we have proposed combining adaptive boosting and active shape model to detect intervertebral discs robustly for automated scan plane planning of spine MRI.

4818
Computer 100
Brain tissue segmentation in fetal MRI using convolutional neural networks with simulated intensity inhomogeneities
Nadieh Khalili1, Nikolas Lessmann1, Elise Turk2,3, Max Viergever1,3, Manon Benders2,3, and Ivana Isgum1,3

1Image Sciences Institute, Utrecht University, Utrecht, The Netherlands, Utrecht, Netherlands, 2Department of Neonatology, Wilhelmina Children’s Hospital, University Medical Center Utrecht, The Netherlands, Utrecht, Netherlands, 3Brain Center Rudolf Magnus, University Medical Center Utrecht, Utrecht, The Netherlands, Utrecht, Netherlands

Automatic brain tissue segmentation in fetal MRI is a challenging task due to artifacts such as intensity inhomogeneity, caused in particular by spontaneous fetal movements during the scan. Unlike methods that estimate the bias field to remove intensity inhomogeneity as a preprocessing step in segmentation process, we aim at improving the robustness of the segmentation method by introducing an intensity inhomogeneity augmentation (IIA). The IIA simulates various patterns of intensity inhomogeneity during the training of the segmentation network. The segmentation results demonstrate an improvement in segmentation performance when the training data is augmented with IIA.


Software & Tools

Exhibition Hall
Thursday 13:45 - 14:45
 Acquisition, Reconstruction & Analysis

4819
Computer 101
SigPy: A Python Package for High Performance Iterative Reconstruction
Frank Ong1 and Michael Lustig1

1University of California, Berkeley, Berkeley, CA, United States

We present SigPy, a Python package designed for high performance iterative reconstruction. Its main features include:

- A unified CPU and GPU Python interface to signal processing functions, including convolution, FFT, NUFFT, wavelet transform, and thresholding functions.

- Convenient classes (Linop, Prox, Alg, App) to build more complicated iterative reconstruction algorithms.

- Commonly used MRI reconstruction methods as Apps, including SENSE, L1-wavelet regularized reconstruction, total-variation regularized reconstruction, and JSENSE.

- MRI-specific functions, including poisson-disc sampling, ESPIRiT calibration, and non-Cartesian preconditioners.

- Simple installation via pip and conda.


4820
Computer 102
BrainQuan: An integrated tool for automated and region-specific analysis of multi-parametric brain MRI data
Xiang Feng1, Meng Xiao Liu2, Guang Yang3, and Xu Yan2

1MR Scientific Marketing, Siemens Healthcare., Beijing, China, 2MR Scientific Marketing, Siemens Healthcare., Shanghai, China, 3Shanghai Key Laboratory of Magnetic Resonance, East China Normal University, Shanghai, China

This abstract presents an integrated tool, namely BrainQuan, developed in Python to automatically segment the brain MRI into sub-regions, align the multi-parametric MRI data into the same space, and then extract the region-specific information from quantitative MRI data, such as quantitative susceptibility maps and diffusion parameters in these brain sub-regions. This tool provides an easy and comprehensive solution for several pilot studies spanning a range of applications: infant brain, brain morphology analysis and neuro-degenerative diseases. BrainQuan might be helpful to establish potential biomarkers from many different quantitative brain MRI data.

4821
Computer 103
DeepRad: An Accessible, Open-source Tool for Deep Learning in Medical Imaging
Jinnian Zhang1, Samuel A Hurley2, Varun Jog1, and Alan B McMillan2

1Electrical & Computer Engineering, University of Wisconsin, Madison, WI, United States, 2Radiology, University of Wisconsin, Madison, WI, United States

Deep learning has shown incredible potential as a powerful tool in medical imaging, however accessibility to deep learning is still limited  for users who lack expertise in computer programming, machine learning, or data science. Existing tools to perform deep learning  have not been designed to be user friendly. We have developed a powerful, flexible, and easy-to-use software specifically tailored to medical imaging for biomedical researchers and physicians with limited programming skills to utilize deep learning for many common tasks.  

4822
Computer 104
Voxel-based morphometry results in first-episode schizophrenia: a comparison of publicly available software packages
Yuanqiang Zhu1, Xingrui Wang1, Xiaocheng Wei2, and Yibin Xi1

1Department of Radiology, Xijing Hospital, Fourth Military Medical University, Xi'an, China, 2GE Healthcare, Xian, China

Investigations of brain structure in schizophrenia using magnetic resonance imaging (MRI) have identified variations in regional grey matter (GM) volume throughout the brain but the results are mixed. This study aims to investigate whether the inconsistent voxel-based morphometry (VBM) findings in schizophrenia are due to the use of different software packages. our data indicate that the GM volume differences between FESZ and HCs depend on which software are used(FSL, SPM), algorithms of GM tissue segmentation and image registration might contribute to these disparate results.

4823
Computer 105
Post Processing Software for Echo Planar Imaging Phase Contrast Sequence
Pan LIU1, Armelle LOKOSSOU1, Sidy FALL1,2, Malek MAKKI1,3, and Olivier BALEDENT1,2

1University of Picardie Jules Verne, CHIMERE EA 7516, Amiens, France, 2CHU-Amiens, Department of Medical Image Processing, Amiens, France, 3CHU-Amiens, MRI Research GIE-FF, Amiens, France

The new sequence Echo Planar Imaging Phase Contrast (EPI-PC) allows real-time imaging of blood flow and can be used to study the effect of breathing unlike to the normal Phase Contrast Magnetic Resonance Imaging sequence (Nor-PC). However, there is no software for the processing of EPI-PC data. We developed new software to visualize, segment and analyze EPI-PC data. We implemented in the software functions as filtering, denoising, segmentation, reconstruction, and extraction that can be applied on EPI-PC signal. This software was easy to use and gave promising results for the quantification of blood flow and the study of breathing effect.

4824
Computer 106
Development of a computer software to quantify bowel motility shown on cine MR imaging by using classical Horn-Schunck approach
Yoshio Kitazume1, Kento Takenaka2, Kazuo Ohtsuka3, and Ukihide Tateishi4

1Diagnostic Radiology, Tokyo Medical and Dental University, Tokyo, Japan, 2Gastroenterology and Hepatology, Tokyo Medical and Dental University, Tokyo, Japan, 3Gastroenterology and Hepatology, Tokyo Medical and Dental Univerisity, Tokyo, Japan, 4Tokyo Medical and Dental University, Tokyo, Japan

We developed a computer software which quantifies the small bowel motility shown on cine MR imaging using optical flow algorithm with Horn-Schunck approach, by adding a preprocessing step for analyzing cine MR images. A high Pearson’s correlation coefficient was obtained between direct measurement on cine MR and motility map value (r= 0.83 [95% confidence interval: 0.83 – 0.95, P<0.0001]). 

4825
Computer 107
High reproducibility and robustness to lesions, but large software and scanner effects for mean upper cervical cord area (MUCCA) measurement in MS
Merlin M Weeda1, Sander M Middelkoop1, Martijn D Steenwijk2, Marita Daams1, Houshang Amiri1, Iman Brouwer1, Joep Killestein3, Bernard MJ Uitdehaag3, Iris Dekker3, Carsten Lukas4, Barbara Bellenberg4, Frederik Barkhof1,5, Petra JW Pouwels1, and Hugo Vrenken1

1Department of Radiology and Nuclear Medicine, MS Center Amsterdam, Amsterdam Neuroscience, Amsterdam UMC - location VUmc, Amsterdam, Netherlands, 2Department of Anatomy and Neurosciences, MS Center Amsterdam, Amsterdam Neuroscience, Amsterdam UMC - location VUmc, Amsterdam, Netherlands, 3Department of Neurology, MS Center Amsterdam, Amsterdam Neuroscience, Amsterdam UMC - location VUmc, Amsterdam, Netherlands, 4Diagnostic and Interventional Radiology and Nuclear Medicine, St. Josef Hospital, Ruhr University, Bochum, Germany, 5Institutes of Neurology and Healthcare Engineering UCL, London, United Kingdom

Atrophy of the spinal cord is known to occur in multiple sclerosis (MS). To measure such atrophy, the mean upper cervical cord area (MUCCA) can be assessed. We tested five different (semi-)automated spinal cord segmentation methods (SCT-PropSeg, SCT-DeepSeg, ITK-SNAP, NeuroQLab, Xinapse JIM) in terms of their reproducibility, robustness, and the influence of lesions on the segmentations. MUCCA from all scanners was highly reproducible within-scanner, but not between-scanner or between-methods. The presence of lesions in the upper cervical cord did not affect the accuracy of MUCCA measurements in any of the methods tested.

4826
Computer 108
Configuring, Viewing, Exploring and Exporting Reproducible, Vendor-Independent MRI Pulse Sequences
Cristoffer Cordes1, Simon Konstandin1, Daniel Mensing1, Saulius Archipovas1, Robin Niklas Wilke1, and Matthias Günther1,2

1MR Physics, Fraunhofer MEVIS, Bremen, Germany, 2MR-Imaging and Spectroscopy, Faculty 01 (Physics/Electrical Engineering), University of Bremen, Bremen, Germany

This work introduces a web technology-based tool that can load device vendor-independent sequence descriptions of a previously described format to then provide interactive tools for configuring protocol parameters, viewing pulse sequence diagrams and details, and exporting raw pulse shapes. 

4827
Computer 109
SEPIA – SuscEptibility mapping PIpeline tool for phAse images
Kwok-Shing Chan1 and José P. Marques1

1Donders Institute for Brain, Cognition and Behaviour, Radboud University, Nijmegen, Netherlands

With the ever-increasing number of quantitative susceptibility mapping (QSM) methods and research applications, it becomes difficult for application-driven researchers to choose a (best) QSM method or pipeline for their study. Here, we present a susceptibility mapping pipeline tool for phase images (SEPIA) which includes a user interface for non-experienced users and the possibility of generating code that can be used for scripting large studies. SEPIA incorporates various QSM toolboxes available in Matlab as well as a wide range of methods to process MR phase data, including signal phase unwrapping, background field removal and field-to-source inversion. 

4828
Computer 110
A Cloud Platform for Longitudinal Follow-up for Patients with Glioblastoma
Saumya Gurbani1, Karthik K Ramesh1, Hyunsuk Shim1, and Brent Weinberg1

1Emory University, Atlanta, GA, United States

Patients diagnosed with glioblastoma are typically treated with a combination of stereotactic surgical resection followed by chemoradiation. Follow-up of these patients post-treatment involves regular imaging to identify disease recurrence and plan adjuvant therapies. In this work, we present a cloud app that will facilitate radiologists and the treating physician team in quantitatively tracking post-treatment disease course using semi-automated segmentation of tumor and a structured scoring system to standardize monitoring of disease progression.

4829
Computer 111
Bloch image simulations of brain pulse sequences using a GPU-installed gaming PC
Katsumi Kose1 and Ryoichi Kose1

1MRI simulations Inc., Tokyo, Japan

Bloch image simulations for typical brain pulse sequences were performed using a GPU-installed gaming laptop PC and a numerical brain phantom. Artifact-free brain MR images were obtained by the Bloch image simulation using optimized numbers of subvoxels. Because the simulation times were the same order as the imaging time for the experiments, we concluded that the Bloch image simulator installed in an inexpensive gaming PC can be a powerful research tool for many MRI engineers and scientists.

4830
Computer 112
New neuroimaging technologies in SPM: BIDS, docker, boutique, and quality control
Tanguy Duval1 and Vincent Lubrano1,2

1UMR 1214 Toulouse Neuroimaging Center, INSERM, Toulouse, France, 2Department of Neurosurgery, University Hospital of Toulouse, Toulouse, France

Today, sharing pipelines across the community is still a complex issue. New technologies and standards, however, change our methods for better collaboration. Here we propose to integrate them seamlessly into the stable and popular SPM pipeline manager. The graphical user interface gives enough flexibility for understanding, modifying, creating and sharing the standard pipelines that are not available today. These pipelines can finally be run with a simple BIDS-app command on any computer.

4831
Computer 113
Application of memory reduced NUFFT to multi-dimensional non-Cartesian MRI
Jyh-Miin Lin1, Grzegorz Kowalik 2, Javier Montalt Tordera1, Benoit Sarthou3, Philippe Ciuciu3, Jennifer Steeden4, and Vivek Muthurangu4,5

1Institute of Cardiovascular Science, University College London, London, United Kingdom, 2University College London, London, United Kingdom, 3CEA/NeuroSpin & INRIA-CEA Parietal team, Gif-sur-Yvette, France, 4Children's Cardiovascular Disease, University College London, London, United Kingdom, 5Great Ormond Street Hospital, London, United Kingdom

A precomputed interpolation matrix on a GPU has been commonly used for fast iterative NUFFT MRI reconstructions, but the size of a 3D interpolation matrix may exceed the memory available on a single GPU. We propose a memory reduced interpolation method that would reduce the size of a multidimensional non-Cartesian interpolation matrix on a GPU. The memory reduced NUFFT reduces the matrix size by more than 90%, while allowing a performance of 38% - 106% of the precomputed version. We also apply the memory reduced NUFFT to a large scale 3D generalized basis pursuit denoising algorithm (GBPDNA) reconstruction.

4832
Computer 114
The qMRLab workflow: From acquisition to publication
Agah Karakuzu1,2, Mathieu Boudreau1,2, Tanguy Duval1, Ilana Leppert 3, Tommy Boshkovski1, Julien Cohen-Adad1,4, and Nikola Stikov1,2

1NeuroPoly Lab, Institute of Biomedical Engineering, Polytechnique Montreal, Montreal, QC, Canada, 2Montreal Heart Institute, Montreal, QC, Canada, 3Montreal Neurological Institute, McGill University, Montreal, QC, Canada, 4Unité de Neuroimagerie Fonctionnelle (UNF), Centre de recherche de l’Institut Universitaire de Gériatrie de Montréal (CRIUGM), Montreal, QC, Canada

qMRLab is open-source software that provides a wide selection of quantitative MRI (qMRI) methods for data fitting, simulation and protocol optimization. Currently, most qMRI methods are developed in-house and are difficult to port across sites. Our vision for qMRLab is to build standardized workflows for these methods, beginning at the scanner console and extending all the way to journal publication. We developed a web portal (https://qmrlab.org) that includes interactive tutorials and Jupyter Notebooks running on BinderHub, tailored for qMRI methods. The last piece of this workflow puzzle is the integration of qMRLab on MR systems, by deploying it as a plugin on a custom MRI application development platform (e.g. RTHawk).

4833
Computer 115
MR Research in the cloud - preliminary results at Columbia University
Can Akgun1 and John Thomas Vaughan2

1Flywheel, Cambridge, MA, United States, 2Columbia University MR Research Center (CMRRC), Columbia University, New York, NY, United States

MR researchers are challenged with managing large data sets, scaling complex computational analyses, and supporting cross-functional collaboration. To tackle these challenges, Columbia University has fully integrated their data management and processing in the cloud to take advantage of high-performance yet low-cost storage, scalable on-demand compute resources, and secure regulatory-compliant infrastructure for sharing of data and algorithms.  The result is a platform that has enabled more efficient workflows, greater productivity, and multi-site collaboration.

4834
Computer 116
A Web-Based Data Management System as a Collaborative Imaging Research Platform
Norman Young1, Jonathan Resnick1, Steven Wranovsky1, Thanh Huynh1, Jasper Yeh1, Stephen Leung1, Stewart Bright1, and Jay Liu1

1Research and Development, Synaptive Medical, Toronto, ON, Canada

A web-based data management system specifically aimed at imaging researchers is presented as a possible solution to the challenges of systematic data management and processing in a research environment.  The system was employed during the development of a head-only MRI for post-processing quality assurance.  Extending the use of the system to facilitate training of machine learning algorithms is proposed.

4835
Computer 117
MRIReco.jl: An Extensible Open-Source Image Reconstruction Framework written in Julia
Tobias Knopp1,2 and Mirco Grosser1,2

1Section for Biomedical Imaging, University Medical Center Hamburg-Eppendorf, Hamburg, Germany, 2Institute for Biomedical Imaging, Hamburg University of Technology, Hamburg, Germany

Image reconstruction plays a major role in the recent years of development in magnetic resonance imaging (MRI) and has been one of the main drivers for reductions in scan time. Within this work we introduce a new software package MRIReco.jl that is very flexible to use and allows for rapid development of new reconstruction algorithms. The package uses the programming language Julia, which is very suitable for implementing reconstruction algorithms on a high abstraction level while still allowing for the generation of runtime-optimized machine code.

4836
Computer 118
FeAture Explorer (FAE): a Tool of Radiomics Feature Analysis and Exploration
Yang Song1, Jing Zhang1, Yu-Dong Zhang2, Xu Yan3, Yida Wang1, Minxiong Zhou4, Bingwen Hu1, and Guang Yang1

1Shanghai Key Laboratory of Magnetic Resonance, East China Normal University, Shanghai, China, 2Department of Radiology, the First Affiliated Hospital with Nanjing Medical University, Nanjing, China, 3MR Scientific Marketing, Siemens Healthcare, Shanghai, China, 4Shanghai University of Medicine & Health Sciences, Shanghai, China

Radiomics studies often requires researchers spend large amount of time trying out various combinations of different data preprocessing strategies, feature selection algorithms, classifiers, and associated hyper-parameters to find the best model. We developed a tool with graphics user interface named FeAture Explorer (FAE) to automate this tedious process. With FAE, to find the best model, researchers only need to specify the choices for each step in radiomics pipeline and let FAE do the rest. Results, such as clinical statistics of each model, can be reviewed and visualized. We used the PROSTATEx dataset to illustrate the function of FAE.

4837
Computer 119
Gadgetron Inline AI: Effective Model inference on MR scanner
Hui Xue1, Rhodri Davies2, David Hansen3, Ethan Tseng4, Marianna Fontana5, James C. Moon2, and Peter Kellman1

1National Heart, Lung and Blood Institute, National Institutes of Health, Bethesda, MD, United States, 2Barts Heart Centre, London, United Kingdom, 3Gradient Software, Skødstrup, Denmark, 4NIH, National Heart, Lung and Blood Institute, Bethesda, MD, United States, 5National Amyloidosis Centre, RoyalFree Hospital, London, United Kingdom

We extended Gadgetron, a widely used open-source framework, to support AI inference on clinical MR scanners. Specially designed software modules (InlineAI) was added to Gadgetron, allowing to load and apply AI neural network models on incoming MR data for compelte "in-line" fashion. That is, without any user interaction, results will be sent back to scanner and available immediately after data acquisition. Two AI based applications were developed as demenstration: Inline AI cine segmenation and perfusion flow mapping and analysis. 

4838
Computer 120
Automated Reconstruction Processing
Eric A. Borisch1, Roger C. Grimm1, and Stephen J. Riederer1

1Radiology, Mayo Clinic, Rochester, MN, United States

A new pair of open-source tools designed to make it easier for researchers to perform automated (no operator intervention) processing of acquired data is described. The first tool handles collection of the required input files on the MR system, submission to an external reconstruction server, and retrieval and import of resulting DICOM images to the system. The second tool manages the reconstruction system, handling prioritization, launching, and monitoring of the reconstruction process.

4839
Computer 121
3D Model-Based Parameter Quantification on Resource Constrained Hardware using Double-Buffering
Oliver Maier1, Matthias Schloegl1, Kristian Bredies2, and Rudolf Stollberger1,3

1Institute of Medical Engineering, Graz University of Technology, Graz, Austria, 2Institute of Mathematics and Scientific Computing, University of Graz, Graz, Austria, 3Biotechmed, Graz, Austria

Reconstructing 3D parameter maps of huge volumes entirely on the GPU is highly desirable due to the offered computation speed-up. However, GPU memory restrictions limit the coverable volume. To overcome this limitation, a double-buffering strategy in combination with model-based parameter quantification and 3D-TGV regularization is proposed. This combination warrants whole volume reconstruction while maintaining the speed advantages of GPU-based computation. In contrast to sequential transfers, double-buffering splits the volume into blocks and overlaps memory transfer and kernel execution concurrently, hiding memory latency. The proposed method is able to reconstruct arbitrary large volumes within 5.3 min/slice, even on a single GPU.

4840
Computer 122
Development of Interpreter Module for Generating Varian VNMRJ Compatible Pulse Sequences using Pulseq Open-Source Toolbox
Courtney Bauer1 and Steven M. Wright1

1Electrical and Computer Engineering, Texas A&M University, College Station, TX, United States

Introduction of a Pulseq interpreter module to enable use of the Pulseq toolbox with Varian-based legacy systems. Targeted as an aid for both educational and research applications, the interpreter module’s development centered around flexibility and ease of use.  Preliminary evaluation of the interpreter module has presented promising results when compared to the existing Varian standard sequences.  The interpreter module is still in refinement, with plans to introduce new features such as variable names for amplitudes and comparison methods to identify user defined shapes that are already present in the system library.

4841
Computer 123
Dynamic platform-independent MRI vs. manufacturer’s implementations
Simon Konstandin1, Cristoffer Cordes1, and Matthias Günther1,2

1MR Physics, Fraunhofer MEVIS, Bremen, Germany, 2MR-Imaging and Spectroscopy, Faculty 01 (Physics/Electrical Engineering), University of Bremen, Bremen, Germany

MR sequence development is usually performed within vendor-specific frameworks, which do not allow for an easy sequence transfer to other manufacturers’ scanners. A platform-independent rapid prototyping environment for MR sequences was presented to allow both, a sequence transfer without code compilation and the generation of dynamic sequences at the scanner. This framework was used to implement a set of standard sequences and modules, which can easily be exchanged or implemented into different sequences. The aim of this work is to show that this approach of vendor-independent sequence development produces same image results as the sequences provided by the manufacturer.

4842
Computer 124
A Realistic Numerical Simulation for Fetal Cardiac MRI
Christopher W. Roy1,2,3, Davide Marini4, William P Segars5, Mike Seed4,6, and Christopher K. Macgowan2,3

1Radiology, Lausanne University Hospital (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland, 2Medical Biophysics, University of Toronto, Toronto, ON, Canada, 3Translational Medicine, The Hospital for Sick Children, Toronto, ON, Canada, 4Pediatric Cardiology, The Hospital for Sick Children, Toronto, ON, Canada, 5Radiology, Duke University Medical Center, Durham, NC, United States, 6Pediatrics and Diagnostic Imaging, University of Toronto, Toronto, ON, Canada

Validating new techniques for fetal CMR is challenging due to random fetal movement that precludes repeat measurements. Consequently, fetal CMR development has been largely performed using physical phantoms or postnatal volunteers. In this work, we present an open-source simulation designed to aide in the development and validation of new approaches for fetal CMR. Our phantom: Fetal XCMR, builds on established methods for simulating MR acquisitions but is tailored toward the dynamic physiology of the fetal heart and body. We present comparisons between the Fetal XCMR phantom and data acquired in utero, resulting in image quality, anatomy, tissue signals and contrast.

4843
Computer 125
FitLike, a software for the analysis of T1 dispersion for Fast Field-Cycling experimentation
Manuel Petit1, Hana Lahrech1, and Lionel Broche2

1Unit 1205 BrainTech Lab, INSERM, Grenoble, France, 2University of Aberdeen, Aberdeen, United Kingdom

Since early 2000 commercial solutions are available to study the dispersion of T1 with the magnetic field strength. This has generated a growing interest in T1 relaxometry study of sample material and a large amount of data to analyse. Yet data analysis for T1 relaxometry is almost entirely done with homemade software, which makes access to the technology difficult and limits the exchanges between research groups. Here we propose a new tool for the analysis of T1 dispersion profiles, software called FitLike that runs with Matlab. 


Machine Learning for Image Enhancement, Quality Assessment & Synthetic Image Generation

Exhibition Hall
Thursday 13:45 - 14:45
 Acquisition, Reconstruction & Analysis

4844
Computer 126
Single image denoising and noise map estimation using random matrix theory
Hong Hsi Lee1,2, Els Fieremans1,2, and Dmitry S Novikov1,2

1Center for Biomedical Imaging, New York University School of Medicine, New York, NY, United States, 2Center for Advanced Imaging Innovation and Research (CAI2R), New York University School of Medicine, New York, NY, United States

Conventional denoising and noise level estimation typically require data redundancy from multiple measurements or prior assumptions, such as smooth image prior or similarity between image patches. Here, we propose a single image denoising algorithm with noise map estimation by identifying the noise-only principle components based on universal properties of random covariance matrices, with the data redundancy created by segmenting data in the Fourier or wavelet domains. The proposed method is applicable to medical and other imaging modalities with spatially-varying noise, and is particularly beneficial to quantitative MRI acquisitions with a limited number of scans. 

4845
Computer 127
3D MRI Denoising with Wasserstein Generative Adversarial Network
Maosong Ran1, Jinrong Hu2, Yang Chen3, Hu Chen1, Huaiqiang Sun4, Jiliu Zhou1, and Yi Zhang1

1College of Computer Science, Sichuan University, Chengdu, China, 2Department of Computer Science, Chengdu University of Information Technology, Chengdu, China, 3Lab of Image Science and Technology, Southeast University, Nanjing, China, 4Department of Radiology, West China Hospital of Sichuan University, Chengdu, China

MR image is easily affected by noise during the high-speed and high-resolution acquisition procedure. To effectively remove the noise and fully explore the potential of latest technique -- deep learning, in this abstract, we propose a novel MRI denoising method based on generative adversarial network. Specifically, to explore the structure similarity among neighboring slices, 3-D configuration are utilized as the basic processing unit. Residual autoencoder, combined with deconvolution operations are introduced into the generator network. The experimental results show that the proposed method achieves superior performance relative to several state-of-art methods in both noise suppression and structure preservation.

4846
Computer 128
Automated slice-to-volume registration between histology and whole-brain post-mortem MRI
Istvan N Huszar1,2, Menuka Pallebage-Gamarallage2, Benjamin C. Tendler1,2, Sean Foxley3, Mattias P. Heinrich4, Martin R. Turner2, Olaf Ansorge2, Karla L. Miller1,2, and Mark Jenkinson1,2

1Wellcome Centre for Integrative Neuroimaging, FMRIB, University of Oxford, Oxford, United Kingdom, 2Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, United Kingdom, 3Department of Radiology, University of Chicago, Chicago, IL, United States, 4Institute of Medical Informatics, Universität zu Lübeck, Lübeck, Germany

Validating MRI data against histological ground truth is essential in the process of devising disease-specific imaging biomarkers that are sensitive to early microstructural changes in neurodegeneration. Current MRI–histology registration techniques are too labour- or resource-intensive to be used in large-scale studies. We introduce an automated pipeline for registering sparsely sampled, small (25x30mm) 2D stained histological images with 3D post-mortem MRI of the whole human brain. Our tests indicate sub-voxel (<0.5 mm) precision using simulated data, and <1 mm precision with real data. Implemented in a new, flexible image registration framework (TIRL), the pipeline is adaptable to various research needs.

4847
Computer 129
A Two-Step Automated Liver MR Images Quality Assessment based on Convolutional Neural Network
Yida Wang1, Yang Song1, Fang Wang2, Zhe Han2, Lei Shi2, Guoliang Shao2, and Guang Yang1

1Shanghai Key Laboratory of Magnetic Resonance, East China Normal University, Shanghai, China, 2Zhejiang Cancer Hospital, Zhejiang, China

We proposed a two-step approach to evaluate automatically liver MR image quality. Firstly, we used a U-Net to segment the liver region. Then image patches were extracted from this region and another CNN was applied to estimate the quality of  each image patch. The quality of the entire image was calculated based on the total percentage of 'bad' image patches in all patches. Receiver operating characteristic curve and confusion matrix were used to evaluate the performance of the proposed method. The performance of our method was comparable to human image readers.

4848
Computer 130
Reinforcement Learning for Automated Reference-free MR Image Quality Assessment
Annika Liebgott1,2, Jianming Yi2, Thomas Küstner1,2,3, Konstantin Nikolaou1, Bin Yang2, and Sergios Gatidis1

1Diagnostic and Interventional Radiology, University Hospital of Tuebingen, Tübingen, Germany, 2Institute of Signal Processing and System Theory, University of Stuttgart, Stuttgart, Germany, 3School of Biomedical Engineering and Imaging Sciences, King's College London/St Thomas' Hospital, London, United Kingdom

Reinforcement learning is a method aiming to model a learner similar to human learning behavior. In this study, we investigate the possibility to utilize this technique to select an optimal feature set for automated reference-free MR image quality assessment. In our proposed setup, we use Q-learning and a random forest classifier to provide feedback to the learner. Moreover, we investigate a combination of multiple reinforcement learning models. Results show that our random-forest-based reinforcement learning setup can achieve higher accuracies than the previously used support vector machines or feature-based deep neural networks combined with traditional feature reduction like PCA.

4849
Computer 131
A deep autoencoder method for image quality assessment
Andre Maximo1, Chitresh Bhushan2, Desmond T.B. Yeo2, and Thomas K Foo2

1GE Healthcare, Rio de Janeiro, Brazil, 2GE Global Research, Niskayuna, NY, United States

We demonstrate a classification approach for MRI image-quality based on deep auto-encoder that can be trained with samples coming from only one class (eg. only good image-quality). This approach is helpful in situations where class-imbalance is unavoidable (i.e. it is easy to obtain a large number of image samples from one class but very difficult to obtain similar number of samples from other class). Our approach shows excellent accuracy in binary classification with AUC of 0.975 in identifying MRI images of good & bad quality in clinical practice from several sites.

4850
Computer 132
Automated Identification of Noise Signal in Spinal DCE-MRI using Independent Component Analysis and Unsupervised Machine Learning
Lucy Wang1, Yi Wang2,3, Murat Alp Oztek4, Nina Mayr4, Simon Lo4, William Yuh4, and Mahmud Mossa-Basha3

1Department of Biomedical Informatics and Medical Education, University of Washington, Seattle, WA, United States, 2Philips Healthcare, Gainsville, FL, United States, 3Radiology, University of Washington, Seattle, WA, United States, 4Radiation Oncology, University of Washington, Seattle, WA, United States

Dynamic Contrast-Enhanced (DCE) MR perfusion has shown early promise in evaluation of spinal metastatic disease and can improve prediction of treatment responses and post-treatment complications. However, spinal DCE-MRI exams frequently suffer from suboptimal image quality due to factors including cerebral spinal fluid (CSF) and vascular pulsation, respiration, bowel motion and patient bulk motion. Independent component analysis has been successfully used as a method to identify and remove motion artifacts from functional MR images. In this work, we combine ICA with an unsupervised machine learning method to automatically identify image components arising from contrast-enhancing tissues and those due to artifacts.

4851
Computer 133
Deep Neural Networks for Motion Estimation in k-space: Applications and Design
Julian Hossbach1,2, Daniel N. Splitthoff2, Melissa Haskell3, Stephen F. Cauley3, Heiko Meyer2, Josef Pfeuffer2, and Andreas Maier1

1Pattern Recognition Lab, Friedrich-Alexander University Erlangen-Nürnberg, Erlangen, Germany, 2Siemens Healthineers AG, Erlangen, Germany, 3Martinos Center for Biomedical Imaging, Charlestown, MA, United States

While image-based motion estimation with Deep Learning has the advantage of an easier comprehension by a human observer, there are benefits to address the issue in k-space, as the distortion only affects echo trains locally; furthermore, Neural Networks can be designed to rely on the intrinsic k-space structure instead of image features. To our knowledge, these advantages have not been exploited so far. We show that fundamental Deep Neural Network techniques can be used for motion estimation in k-space, by examining different networks and hyperparameters on a simplified problem. We find suitable architectures for extracting 2D transformation parameters from under-sampled k-spaces for slice registration. This leads to a minimum residual of around 1.2 px/deg.

4852
Computer 134
Deep Residual Neural Networks for QSM Background Removal
Juan Liu1, Andrew Nencka1,2, and Kevin Koch1,2

1Joint Department of Biomedical Engineering, Marquette University and Medical College of Wisconsin, Milwaukee, WI, United States, 2Department of Radiology, Medical College of Wisconsin, Milwaukee, WI, United States

Quantitative Susceptibility Mapping (QSM) is a MR post-processing technique that estimates underlying tissue magnetic susceptibilities. In QSM processing pipelines, background field removal is of vital importance to obtain local tissue field estimates for precise susceptibility quantification. Existing background field removal methods such as SHARP, RESHARP, PDF, and LBV can effectively remove the background field. However, they struggled in clinical applications with large slice thickness and resulting non-isotropic resolutions. To address the limitations of these existing pre-processing methods in clinical QSM practice, a deep-learning-based method was proposed to approximate the underlying tissue field maps from total field maps. In-vivo datasets acquired using clinical SWI protocol demonstrated the improved performance of this approach, compared to conventional existing methods.

4853
Computer 135
3T to 7T MRI Synthesis via Deep Learning in Spatial-Wavelet Domains
Liangqiong Qu1, Shuai Wang 1, Yongqin Zhang2, Pew-Thian Yap1, and Dinggang Shen1

1Department of Radiology and BRIC,University of North Carolina at Chapel Hill, Chapel Hill, NC, United States, 2School of Information Science and Technology, Northwest University, Xi'an, China

Ultra-high field 7T MRI scanners, while producing images with exceptional anatomical details, are cost prohibitive and hence highly inaccessible. In this abstract, we propose a novel deep learning network to synthesize 7T T1-weighted images from their 3T counterparts. Our network jointly considers both spatial and wavelet domains to facilitate learning for coarse to fine details. 



4854
Computer 136
Overcoming the Rician Noise Bias of T2* Relaxometry with an Artificial Neural Network (ANN)
Ferdinand Schweser1,2, Thomas Jochmann3, and Robert Zivadinov1,2

1Buffalo Neuroimaging Analysis Center, Department of Neurology, Jacobs School of Medicine and Biomedical Sciences, University at Buffalo, The State University of New York, Buffalo, NY, United States, 2Center for Biomedical Imaging, Clinical and Translational Science Institute, University at Buffalo, The State University of New York, Buffalo, NY, United States, 3Department of Computer Science and Automation, Technische Universität Ilmenau, Ilmenau, Germany

Rician noise represents the major source of bias in parametric fitting techniques, such as the estimation of the T2* relaxation time. This bias is particularly strong when the signal-to-noise ratio is low or T2* values are short, such as in clinical cases of severe brain or liver iron overload.  In this work, we trained a deep convolutional neural network to recognize Rician noise and compute unbiased relaxation parameters from multi-echo gradient echo data.

4855
Computer 137
k-space deep learning for MR herringbone artifact correction
Juyoung Lee1 and Jong Chul Ye1

1KAIST, Daejeon, Korea, Republic of

Herringbone artifact is caused by power fluctuation of MR equipment or unstable shielding. Herringbone artifact image is difficult to analyze because it scatters on whole image region of single or multiple slices. There is a study for MR artifact correction which can be represented as sparse outliers on k-space. This method exploits the duality between the low-rankness of Hankel matrix in k-space and the sparsity in the image domain. However, this method has high computational complexity, and consumes much time. In this research, we suggest the new effective and fast MR artifact correction method using deep learning.

4856
Computer 138
Gibbs-Ringing Artifact reduction in MR images with varying sampling levels Via a Single Convolutional Neural Network
Guohui Ruan1, Qianqian Zhang1, Biaoshui Liu2, Wei Yang1, Yingjie Mei3, Ed X. Wu4, and Yanqiu Feng1

1Guangdong Provincial Key Laboratory of Medical Image Processing & Key Laboratory of Mental Health of the Ministry of Education, School of Biomedical Engineering, Southern Medical University, Guangzhou, China, Guangzhou, China, 2Sun Yat-Sen University Cancer Center, Guangzhou, China, 3Philips Healthcare, Guangzhou, China, Guangzhou, China, 4Department of Electrical and Electronic Engineering, The University of Hong Kong, Hong Kong SAR, China, Laboratory of Biomedical Imaging and Signal Processing, The University of Hong Kong, Hong Kong SAR, China, Hong Kong, China

Gibbs-ringing artifact is caused by the insufficient sampling of the high frequency data. And in clinical practice, the appearance of ringing artifact, i.e. the real sampling level, is not accurately obtained. To address this problem, a single convolutional neural network (CNN) has been trained for reducing Gibbs-ringing artifact in MR images under varying sampling levels. The experimental results demonstrate that Gibbs-ringing artifact can be effectively reduced by the proposed method without introducing noticeable blurring. 

4857
Computer 139
Motion Correction of Magnitude MR Images using Generative Adversarial Networks
Yuan Bian1, Ye Wang2, and Stanley Reeves1

1Electrical and Computer Engineering, Auburn University, Auburn, AL, United States, 2Computer Science and Software Engineering, Auburn University, Auburn, AL, United States

Motion during MRI scan can reduce image quality due to the induced artifacts. We present a novel data-driven motion correction method for magnitude MR images using generative adversarial networks (GANs). GANs (Pix2pix model) is implemented to reduce motion artifacts and reconstruct motion-corrupted images through adversarial training between generator and discriminator to force motion-corrected image close to the reference image. The training set is made of image pairs, which consist of motionless reference images and corresponding motion-simulated images. The proposed method was validated by a simulated motion test set and a real motion (experimental) test set.

4858
Computer 140
Fetal Motion Prediction from Volumetric MRI using Machine Learning
Junshen of Xu1, Molin Zhang2, Larry Zhang1,3, Ellen Grant4,5, Polina Golland1,3, and Elfar Adalsteinsson1,6

1Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States, 2Department of Engineering Physics, Tsinghua University, Beijing, China, 3Computer Science and Artificial Intelligence Laboratory (CSAIL), Massachusetts Institute of Technology, Cambridge, MA, United States, 4Harvard Medical School, Boston, MA, United States, 5Fetal-Neonatal Neuroimaging and Developmental Science Center, Boston Children’s Hospital, Boston, MA, United States, 6Institute for Medical Engineering and Science, Massachusetts Institute of Technology, Cambridge, MA, United States

Prospective motion correction is a challenge in clinical fetal MR imaging as fetal motion is erratic and often substantial. To address this problem, we propose a two-stage machine learning pipeline to extract fetal poses from echo planar MRI volumes at previous time points to predict future pose. This pipeline can be used to learn kinematic models of fetal motion and serve as valuable auxiliary information for real-time, online slice prescription in fetal MRI.

4859
Computer 141
Head Movement Detection from Radial k-Space Lines using Convolutional Neural Networks – A Digital Phantom Study
Maximilian Wattenberg1, Jannis Hagenah2, Constantin Schareck1, Floris Ernst2, and Martin A. Koch1

1Institute of Medical Engineering, Universität zu Lübeck, Lübeck, Germany, 2Institute for Robotics and Cognitive Systems, Universität zu Lübeck, Lübeck, Germany

Magnetic resonance imaging-guided linear particle accelerators use reconstructed images to adapt the radiation beam to the tumor location. Image-based approaches are relatively slow, causing healthy tissue to be irradiated upon subject movement. This study targets on the use of convolutional neural networks to estimate rigid patient movements directly from few acquired radial k-space lines. Thus, abrupt patient movements were simulated in image data of a head. Depending on the number of acquired spokes after movement, the network quantified this motion precisely. These first results suggest that neural network-based navigators can help accelerating beam guidance in radiotherapy.

4860
Computer 142
Deep Learning based Velocity Aliasing Correction for 4D Flow MRI
Haben Berhane1, Hassan Haji-Valizadeh2, Joshua Robinson1, Michael Markl2, and Cynthia Rigsby1

1Lurie Childrens Hospital of Chicago, Chicago, IL, United States, 2Northwestern University, Chicago, IL, United States

We developed a convolutional neural network to detect and correct velocity aliasing in 4D Flow datasets. Our network uses an Unet architecture and was trained, validated, and tested on 100, 10, and 100 datasets respectively. It was able to detect as many or more phase wrapped voxels compared to the conventional algorithm and performed better on highly aliased regions of the dataset.

4861
Computer 143
Fetal Pose Estimation via Deep Neural Network by Detection of Fetal Joints, Eyes, and Bladder
Molin Zhang1, Junshen Xu2, Esra Turk3, Larry Zhang2,4, P.Ellen Grant3,4, Karen Ying1, Polina Golland2,4, and Elfar Adalsteinsson2,5

1Department of Engineering Physics, Tsinghua university, BeiJing, China, 2Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, United States, 3Harvard Medical School, Boston, MA, United States, 4Fetal-Neonatal Neuroimaging and Developmental Science Center, Boston Children’s Hospital, Boston, MA, United States, 5Institute for Medical Engineering and Science, Massachusetts Institute of Technology, Cambridge, MA, United States

Neural networks and deep learning have achieved great success in human pose estimation through the identification of key human points in conventional photography and video. We propose fetal pose estimation in a time series of echo planar MRI volumes of the pregnant abdomen via deep learning algorithms for detection of key fetal landmarks, including joints, eyes, and bladder. Fetal pose estimation in an EPI time series yields novel means of quantifying fetal movements in health and disease, and enables learning of kinematic models that may enhance mitigation of fetal motion artifacts during MRI acquisition.

4862
Computer 144
Improving the Quality of Synthetic FLAIR Images with Deep Learning Using a Conditional Generative Adversarial Network for Pixel-by-Pixel Image Translation
Akifumi Hagiwara1, Yujiro Otsuka2, Masaaki Hori2, Yasuhiko Tachibana3, Kazumasa Yokoyama4, Shohei Fujita2, Christina Andica2, Koji Kamagata2, Ryusuke Irie2, Saori Koshino1, Tomoko Maekawa2, Lydia Chougar5, Akihiko Wada2, Mariko Yoshida Takemura2, Nobutaka Hattori4, and Shigeki Aoki2

1Radiology, The University of Tokyo Hospital, Tokyo, Japan, 2Radiology, Juntendo University Hospital, Tokyo, Japan, 3Radiology, National Institute of Radiological Sciences, Chiba, Japan, 4Neurology, Juntendo University Hospital, Tokyo, Japan, 5Radiology, Hopital Saint-Joseph, Paris, France

Synthetic FLAIR images are of lower quality than conventional FLAIR images. Here, we aimed to improve the synthetic FLAIR image quality using deep learning with pixel-by-pixel translation through conditional generative adversarial network training. Forty patients with MS were prospectively included and scanned to acquire synthetic MRI and conventional FLAIR images. Acquired data were divided into 30 training and 10 test datasets. Using deep learning, we improved the synthetic FLAIR image quality by generating FLAIR images that have contrast that is closer to that of conventional FLAIR images and fewer granular and swelling artifacts, while preserving the lesion contrast.

4863
Computer 145
Synthesizing T2 Maps from Morphological OAI Scans Using Conditional GANs and a Split U-Net
Bragi Sveinsson1,2,3, Bo Zhu1,2,3, Akshay Chaudhari4, Neha Koonjoo1,2,3, and Matthew Rosen1,2,3

1Martinos Center, Massachusetts General Hospital, Charlestown, MA, United States, 2Harvard Medical School, Boston, MA, United States, 3Physics, Harvard University, Cambridge, MA, United States, 4Stanford University, Stanford, CA, United States

We explore the feasibility of artificially adding an exam to an MRI scan protocol by synthesizing the desired exam from the acquired images. To achieve this, we both use a normal U-Net as well as a modified U-Net structure, which takes advantage of prior information of which exams of the protocol are most relevant to the high-resolution and low-resolution components of the desired contrast. We demonstrate results based on synthesizing T2 relaxation time maps using imaging data obtained from the Osteoarthritis Initiative.

4864
Computer 146
Build-a-FLAIR: The synthetic generation of T2-FLAIR contrast from T2-weighted and diffusion metric images through a deep neural network.
Andrew S. Nencka1, Brad J. Swearingen2, Kevin M Koch1, and Michael McCrea2

1Radiology, Medical College of Wisconsin, Milwaukee, WI, United States, 2Neurosurgery, Medical College of Wisconsin, Milwaukee, WI, United States

A deep neural network is presented to synthetically generate T2FLAIR weighted images from other standard neuroimaging acquisitions. Network performance improved with input images that share components with similar physical sources of contrast as the T2FLAIR contrast, while performance was degraded when disparate sources of contrast, like fractional anisotropy, were included. This suggests that a level of feature engineering is appropriate when building deep neural networks to perform style transforms with respect to MRI contrast, with input features containing shared physical sources of contrast with the desired output contrast. In the optimally trained network, pathology present in the acquired T2FLAIR images and not present in the training dataset was correctly reconstructed.

4865
Computer 147
The impact of variable MRI acquisition parameters on deep learning-based synthetic CT generation
Mateusz C. Florkow1, Frank Zijlstra1, Koen Willemsen2, René M. Castelein2, Peter R. Seevinck1,3, and Marijn van Stralen1,3

1Image Sciences Institute, University Medical Center Utrecht, Utrecht, Netherlands, 2Department of Orthopedics, University Medical Center Utrecht, Utrecht, Netherlands, 3MRIguidance, Utrecht, Netherlands

Deep learning-based synthetic CT generation models are generally trained and evaluated on MR images obtained with a single set of acquisition parameters. In this study, we investigated the robustness of such models to clinically plausible changes in acquisition parameters by training and evaluating models on MR images acquired and reconstructed from gradient echo sequences at different echotimes (TE), resolution and flip angles. We investigated the sensitivity to TEs by training models on randomly interspersed multi-echo gradient echo MR images acquired at different TEs. Multi-echo trained models achieved better generalization performance to varying acquisition parameters without excessively compromising results on dedicated data.

4866
Computer 148
Visualizing the “ideal” input MRI for synthetic CT generation with a trained deep convolutional neural network: Can we improve the inputs for deep learning models?
Andrew P. Leynes1,2 and Peder E.Z. Larson1,2

1University of California San Francisco, San Francisco, CA, United States, 2UC Berkeley - UC San Francisco Joint Graduate Program in Bioengineering, Berkeley and San Francisco, CA, United States

Deep learning has found wide application in medical image reconstruction, transformation, and analysis tasks. Unlike typical machine learning workflows, MRI researchers are able to change the characteristics of images that are used as inputs to deep learning models. We proposed an algorithm that allows us to visualize the “ideal” input images that would provide the least error for a trained deep neural network. We apply this visualization technique on a deep convolutional neural network that converts Dixon MRI to synthetic CT images. We briefly characterize the optimization behavior and qualitatively analyze the features of the “ideal” input image.

4867
Computer 149
Synthetic MRI with water suppression technique  to reduce CSF partial-volume artifacts
Tokunori Kimura1, Yuki Takai2, Hiroshi Kusahara2, Hitoshi Kanazawa2, and Ryo Shiroishi3

1Department of Radiological Sciences, Shizuoka College of Medicalcare Science, Hamamatsu, Japan, 2MRI development department, Canon Medical Systems corp., Otawara, Japan, 3Clinical Research and Development Center, Canon Medical Systems corp., Otawara, Japan

We proposed a new synthetic-MRI technique combined with water suppression to reduce CSF partial volume effects (PVE) artifacts problematic in a conventional synthetic-MRI. Our water suppression was simply achieved by subtracting additionally acquired long-TE SE image of water signal dominant. After the quantitative parameter maps of original and with water suppression were generated, water-suppressed synthetic-SE and -FLAIR images were calculated using those suitable combinations. We demonstrated that CSF PVE artifacts were dramatically reduced in our proposed synthetic-FLAIR, and furthermore that, by the two-compartment model simulation and volunteer MR brain study, our synthetic-SE provided better gray-white matter contrasts compared to our synthetic-FLAIR.

4868
Computer 150
Towards Contrast-Independent Automated Motion Detection Using 2D Adversarial DenseNets
Silvia Arroyo-Camejo1,2, Benjamin Odry1, Xiao Chen1, Kambiz Nael3, Luoluo Liu1,4, David Grodzki1,2, and Mariappan S. Nadar1

1Digital Technology and Innovation, Siemens Healthineers, Princeton, NJ, United States, 2Magnetic Resonance, Siemens Healthcare GmbH, Erlangen, Germany, 3Radiology, Icahn School of Medicine at Mount Sinai, New York City, NY, United States, 4Electrical Engineering, Johns Hopkins University, Baltimore, MD, United States

Patient motion is a challenging and common source of artifacts in MRI. Two recent studies investigating motion detection with convolutional neural networks showed promising results, but did not generalize to varying MRI contrasts. We present a unified, domain adapted deep learning routine to provide automated image motion assessment in MR brain scans with T1 and T2 contrast. We aim to limit the influence of varying image contrasts, scanner models, and scan parameters in the motion detection routine by using adversarial training.


Machine Learning for Prediction & Image Analysis

Exhibition Hall
Thursday 13:45 - 14:45
 Acquisition, Reconstruction & Analysis

4869
Computer 151
Feature Reduction and Selection: a Study on their Importance in the Context of Radiomics
Annika Liebgott1,2, Janik Steyer-Ege2, Tobias Hepp1, Thomas Küstner1,2,3, Konstantin Nikolaou1, Bin Yang2, and Sergios Gatidis1

1Diagnostic and Interventional Radiology, University Hospital of Tuebingen, Tübingen, Germany, 2Institute of Signal Processing and System Theory, University of Stuttgart, Stuttgart, Germany, 3School of Biomedical Engineering and Imaging Sciences, King's College London/St Thomas' Hospital, London, United Kingdom

Using large amounts of image features in the context of Radiomics to perform complex image analysis tasks yields promising results for clinical applications. While it is easy to extract a large amount of features from medical images, it is complex to select the right features for a specific scientific problem. This study aims to show, how important it is to pay attention to choosing the right technique to select the most suitable features by means of feature reduction or selection on the example of two Radiomics-related MR image classification tasks.

4870
Computer 152
Preoperative prediction of pelvic lymph nodes metastasis in early-stage cervical cancer using radiomics model based on T2-weighted and diffusion-weighted MRI
Tao Wang1,2, Tingting Gao3, Liyu Huang3, and Ming Zhang2

1Department of Radiology, Shaanxi Provincial People's Hospital, xi'an, China, 2Department of Medical Imaging, First Affiliated Hospital of Xi’an Jiaotong University, Xi’an, China, 3School of Life Science and Technology, Xidian University, Xi’an, China

1.       A radiomics nomogram for preoperatively predicting of PLN metastasis in patients with ECC was developed and validated.

2.       The model displayed good performance (C-index=0.893 in primary cohort and C-index=0.922 in validation cohort).

3.       The radiomics nomogram showed a significant improvement over the clinical nomogram in predicting PLN metastasis.

4.       The radiomics signature derived from the combined T2WI and DWI has the best performance.


4871
Computer 153
Diagnosis of Multiple Sclerosis Subtype through Machine Learning Analysis of Frontal Cortex Metabolite Profiles
Abhinav V. Kurada1, Kelley M. Swanberg1,2, Hetty Prinsen2, and Christoph Juchem1,2,3,4

1Biomedical Engineering, Columbia University School of Engineering and Applied Science, New York, NY, United States, 2Radiology and Biomedical Imaging, Yale University School of Medicine, New Haven, CT, United States, 3Neurology, Yale University School of Medicine, New Haven, CT, United States, 4Radiology, Columbia University Medical Center, New York, NY, United States

The onset and progression of multiple sclerosis (MS) is accompanied by changes in brain biochemistry. Magnetic resonance spectroscopy (MRS) is a powerful tool for investigating these changes in vivo. Machine learning analysis of MRS-derived biochemical profiles may reveal metabolic patterns inherent in certain MS subtypes to inform their diagnosis. By employing a feature set of only metabolite concentrations derived from brain MRS data acquired at 7 Tesla, we achieved an 80% validation set accuracy for differentiating MS patients from healthy controls and a 70% validation set accuracy for differentiating relapsing-remitting and progressive MS patients. 

4872
Computer 154
Differentiation of Osteosarcoma and Ewing Sarcoma Using Radiomic AnalysisBased on T2 and CET1 MRI
Yi Dai1,2, Nan Hong2, and Guanxun Cheng1

1Peking University Shenzhen Hospital, Shenzhen, China, 2Peking University People's Hospital, Beijing, China

In this study, we assessed the ability of our newly established radiomic model based on using multiparametric MR data to help differentiate OS from EWS of the pelvis. We evaluated 16 features that were extracted and selected by using the LASSO method. Our radiomics model yielded favorable results and constituted a new technique for the discrimination of OS and EWS. The AUC was high for both T2-FS and CET1. High specificity was achieved when using data both from T2-FS and CET1 (82.9% and 100%, respectively) and the sensitivity was also high from T2-FS (74.2%). In brief, we believe that the methodology developed in this work may serve as a reliable additional tool for differentiation OS from EWS.

4873
Computer 155
Generated data can boost the recognition performance for Intervertebral disc herniation
Fei Gao1, Shui Liu2, Xiaodong Zhang2, Jue Zhang1,3, and Xiaoying Wang2,3

1College of Engineering, Peking University, Beijing, China, 2Department of Radiology, Peking University First Hospital, Beijing, China, 3Academy for Advanced Interdisciplinary Studies, Peking University, Beijing, China

Although deep convolutional neural network has shown encouraging performance regarding lesion classification, it is limited due to the high requirement of data labeling. In this study, we attempted to improve the recognition performance under limited labeled data using generated data for lumbar intervertebral disc herniation classification.

4874
Computer 156
Deep Predictive Modeling of Dynamic Contrast-Enhanced MRI Data
Jiacheng Jason He1, Christopher Sandino1, Shreyas Vasanawala1,2, and Joseph Cheng2

1Electrical Engineering, Stanford University, Stanford, CA, United States, 2Radiology, Stanford University, Stanford, CA, United States

This work demonstrates the use of recurrent generative spatiotemporal autoencoders to predict up to fifteen future frames of abdominal DCE-MRI video data, starting with only three ground truth input frames for context. The objective is to predict what healthy patient video data and organ-specific contrast curves look like, to expedite anomaly detection and enable pulse sequence optimization. The model in this study shows promise; it was able to learn contrast changes without losing structural resolution during training time, and lays the foundation for future work.

4875
Computer 157
Automatic detection of age- and sex-related differences in human brain morphology
Renzo Phellan1, Lívia Rodrigues2, Gustavo Retuci Pinheiro2, Andrés Quiroga Soto3,4, Igor Duarte Rodrigues5, Leticia Rittner2, Ricardo Ferrari6, Matthew R G Brown7, Nils D Forkert8, Roberto Medeiros9, and Mariana Bento9

1Department of Radiology and Hotchkiss Brain Institute, University of Calgary, Calgary, AB, Canada, 2School of Electrical and Computer Engineering, University of Campinas, Campinas, Brazil, 3Institute of Physics IFGW, University of Campinas, Campinas, Brazil, 4Brazilian Institute of Neuroscience and Neurotechnology, University of Campinas, Campinas, Brazil, 5Institute of Systems Engineering and Information Technology, Universidade Federal de Itajubá, Itajubá, Brazil, 6Department of Computer Science, Universidade Federal de São Carlos, São Carlos, Brazil, 7Department of Computing Science, University of Alberta, Edmonton, AB, Canada, 8Department of Radiology, University of Calgary, Calgary, AB, Canada, 9Seaman Family Centre, University of Calgary, Calgary, AB, Canada

Research on neurological and mental disorders has shown the diagnostic potential of volumetric brain analysis, also evidencing differences of human brain structures regarding sex and aging in normal subjects. This study aims at identifying the most important volumetric sex- and age-related differences of brain structures using machine learning approaches. It was found that the most important brain structures were different for age- and sex-related differences, which should be taking into account when diagnosing neurological and mental disorders based on morphological features.

4876
Computer 158
Survival prediction from DCE-MRI kinetic parameters in patients with osteosarcoma using deep learning
Junyu Guo1 and Wilburn E. Reddick1

1St Jude Children's Research Hospital, Memphis, TN, United States

DCE-MRI may be a prognostic biomarker for some tumors including osteosarcoma. The purpose of this study was to assess whether a DCE-MRI kinetic parameter map of osteosarcoma can provide prognostic indicators for clinical results using three deep convolution neural networks (DCNN). In this study, we found that DCNNs can provide biomarkers for overall survivals with accuracy over 0.8; three DCNNs have the comparable performance in prediction of clinical results; and the predictions using DCNN with tumor mask were significantly better than those without using tumor mask.

4877
Computer 159
Deep neural network processing of original DCE-MRI data for survival prediction
Junyu Guo1 and Wilburn E. Reddick1

1St Jude Children's Research Hospital, Memphis, TN, United States

DCE-MRI is a valuable tools in many clinical applications, but data analysis is complex. The purpose of this study was to assess whether the original DCE images without complex modeling can be used to predict the clinical results of osteosarcoma using deep convolution neural network (DCNN). We also assess whether the prediction from original images were different from those using the kinetic parameters. We found that DCNN can predict overall survivals with an accuracy of about 0.8 using a set of 2D DCE tumor images, which is not significantly different from results based on kinetic parameter maps.

4878
Computer 160
Characterizing MRI Biomarkers for Conversion Prediction of Preclinical Mild Cognitive Impairment
Yongsheng Pan1,2, Mingxia Liu*2, Chunfeng Lian2, Ling Yue3, Shifu Xiao3, Yong Xia*1, and Dinggang Shen*2

1School of Computer Science, Northwestern Polytechnical University, Xi'an, China, 2Department of Radiology and BRIC, University of North Carolina at Chapel Hill, Chapel Hill, NC, United States, 3School of Medicine, Shanghai Mental Health Center, Shanghai Jiao Tong University, Shanghai, China

Identifying subjects at the stage of preclinical mild cognitive impairment (pre-MCI) is fundamental for early intervention of pathologic cognitive decline. This study aims to investigate the progression from cognitive normal (CN) and subjective cognitive decline (SCD) to MCI, by characterizing imaging biomarkers in brain MRI data via a deep-learning framework. This deep-learning framework is designed to first evaluate the discriminative capability of regions-of-interest (ROIs) in brain MR images, and then to predict the progression of CN/SCD subjects to MCI within 36 months. The results suggest that brain structure changes at the pre-MCI stage can be objectively detected in MR images by our method.

4879
Computer 161
Detection of White Matter Hyperintensities using Ensemble 3D Deep Learning Networks
Lavanya Umapathy1, Gloria Guzman2, Jose Rosado-Toro2, Gokhan Kuyumcu2, Maria Altbach2, Blair Winegar2, Craig Weinkauf3, and Ali Bilgin1,4

1Electrical and Computer Engineering, University of Arizona, Tucson, AZ, United States, 2Department of Medical Imaging, University of Arizona, Tucson, AZ, United States, 3Department of Surgery, University of Arizona, Tucson, AZ, United States, 4Biomedical Engineering, University of Arizona, Tucson, AZ, United States

White matter hyperintensities (WMH), hyperintense on T2-weighted FLAIR images are prominent features of demyelination and axonal degeneration in cerebral white matter. The time-consuming nature of manual segmentation necessitates the need for faster and reliable automated segmentation algorithms. In this work, we propose three deep learning architectures for WMH detection on 3D FLAIR images: a modified UNET3D, Res-UNET3D and their ensemble combination. Two UNET3D and two Res-UNET3D were trained with random initialization using 3D patches sampled from within the brain. The posterior probabilities for WMH from individual networks were averaged to obtain a revised posterior probability for the ensemble. Performance of the individual networks as well as that of the ensemble was assessed using dice and precision scores.

It was observed that the ensemble of 3D networks yields improved dice and precision scores in comparison to an average of individual networks, thereby reducing the effect of choice of network or parameters. Furthermore, the average dice scores for the ensemble approached the inter-observer variability of human observers.


4880
Computer 162
3D Convolutional Networks to predict Total Knee Replacement using Structural MRI
Tianyu Wang1, Kevin Leung2,3, Kyunghyun Cho1,3, Gregory Chang4,5, and Cem M. Deniz4,6,7

1Center for Data Science, New York University, New York, NY, United States, 2Leonard N. Stern School of Business, New York University, New York, NY, United States, 3Courant Institute of Mathematical Sciences, New York University, New York, NY, United States, 4Department of Radiology, New York University Langone Medical Center, New York, NY, United States, 5Center for Musculoskeletal Care, New York University Langone Medical Center, New York, NY, United States, 6Center for Advanced Imaging Innovation and Research (CAI2R) and Bernard and Irene Schwartz Center for Biomedical Imaging, New York University Langone Medical Center, New York, NY, United States, 7The Sackler Institute of Graduate Biomedical Sciences, New York University School of Medicine, New York, NY, United States

Osteoarthritis (OA) is a chronic degenerative disorder of joints and is the most common reason leading to total knee joint replacement (TKR). In this work, we developed an automated OA-relevant imaging biomarker identification system based on MR images and deep learning (DL) methods to predict knee OA progression. Our results indicate that the combination of multiple MR images with different contrast and resolution provides the best model to predict TKR with AUC 0.88±0.01.

4881
Computer 163
Whole-Brain R1 mapping predicts occupational Mn air exposure: a support vector machine approach
David A Edmondson1,2, Sébastien Hélie3, and Ulrike Dydak1,2

1School of Health Sciences, Purdue University, West Lafayette, IN, United States, 2Department of Radiology and Imaging Sciences, Indiana University School of Medicine, Indianapolis, IN, United States, 3Department of Psychological Sciences, Purdue University, West Lafayette, IN, United States

Manganese (Mn) is a neurotoxin that can lead to symptoms similar to Parkinson’s disease. Welders exposed to welding fume can accumulate quantities of Mn in their brain eliciting T1 contrast effects. Mn exposure estimates are useful for determining a welder’s risk for symptoms, but quantifying Mn in the brain would be more beneficial. While R1 (1/T1) is proportional to local Mn accumulation, the relationship is likely non-linear, complicating interpretation of R1. Therefore, we propose a support vector machine model using whole-brain R1 maps to predict classes as determined by group, Mn air exposure, and excess brain Mn accumulation. 

4882
Computer 164
To evaluate the role of machine learning for characterization of breast lesion using multi-parametric MRI.
Snekha Thakran1, Dinil Sasi S1, Rupsa Bhattacharjee1, Ayan Debnath1, Rakesh Kumar Gupta2, and Anup Singh1,3

1Centre for Biomedical Engineering, Indian Institute of Technology Delhi, New Delhi, India, 2Department of Radiology, Fortis Memorial Research Institute, Haryana, Gurgaon, India, 3Department of Biomedical Engineering, All India Institute of Medical Science, New Delhi, India

The role of machine learning in medical imaging is increasing day by day. It can help in combining a variety of complementary information obtained using multi-parametric MRI(mpMRI). The objective of this study was to differentiate benign vs. malignant breast tumor using machine learning with optimized feature set obtained from mpMRI data. The study included mpMRI data of 49 patients with breast cancer. Quantitative mpMRI parameters as well as texture features were used as feature set in machine learning. The combination of the wrapper method with SVM resulted in high sensitivity (100%) and specificity (93.75%) in the binary classification of benign and malignant.

4883
Computer 165
Differential diagnosis of multiple sclerosis based on the central vein sign assessment using deep learning: a multicentre study.
Mário João Fartaria1,2,3, Jonas Richiardi1,2,3, Pietro Maggi4, Pascal Sati5, Daniel S. Reich5, Cristina Granziera6,7, Meritxell Bach Cuadra2,3,8, and Tobias Kober1,2,3

1Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, 2Department of Radiology, Centre Hospitalier Universitaire Vaudois (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland, 3Signal Processing Laboratory (LTS 5), Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, 4Departement of Neurology, Centre Hospitalier Universitaire Vaudois (CHUV) and University of Lausanne (UNIL), Lausanne, Switzerland, 5Translational Neuroradiology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health (NIH), Bethesda, MD, United States, 6Neurologic Clinic and Policlinic, Departments of Medicine, Clinical Research and Biomedical Engineering, University Hospital Basel and University of Basel, Basel, Switzerland, 7Translational Imaging in Neurology (ThINk) Basel, Department of Medicine and Biomedical Engineering, University Hospital Basel and University of Basel, Basel, Switzerland, 8Medical Image Analysis Laboratory (MIAL), Centre d'Imagerie BioMédicale (CIBM), Lausanne, Switzerland

Prospective multicentre studies are needed to establish the clinical value of the central vein sign for diagnosis of multiple sclerosis. This type of studies requires manual segmentation and classification of lesions with and without the central vein sign, which are time-consuming tasks. In this work, we evaluate the performance of an in-house deep-learning-based prototype algorithm for automated assessment of the central vein sign using data from two different healthcare units.          

4884
Computer 166
Structural- and Functional-Connectivity Convolution Neural Networks (SCFCnn) for Integrated Brain-Behavior Prediction in the HCP dataset
Ying-Chia Lin1,2, Steven Baete1,2, Xiuyuan Wang1,2, and Fernando Boada1,2

1Center for Advanced Imaging Innovation and Research (CAI2R), NYU School of Medicine, New York, NY, United States, 2Center for Biomedical Imaging, Department of Radiology, NYU School of Medicine, New York, NY, United States

In this work, we investigate an efficient structural (SC)- and functional (FC)-connectivity convolution neural network (SCFCnn) architecture applied on both FC and SC to detect the links between individual non-imaging language traits and invivo MRI measurements in a subset of the Human Connectome Project (HCP) s900 dataset. The identified structure-function relationships can be used to infer neurocognitive measures from neuroimaging. Our new architecture outperforms popular deep learning neural networks, confirming the importance of convolutional neural networks applied to brain connectivity for better predictive performance in neurocognitive measurements.

4885
Computer 167
Enhance One-minute EPIMix Brain MRI Exams with Unsupervised Cycle-Consistent Generative Adversarial Network
Jiang Liu1, Enhao Gong2,3, Stefan Skare4, and Greg Zaharchuk2

1Tsinghua University, Beijing, China, 2Stanford University, Stanford, CA, United States, 3Subtle Medical Inc., Menlo Park, CA, United States, 4Karolinska Institutet, Stockholm, Sweden

Recently, a new one-minute multi-contrast echo-planar imaging (EPI) based sequence (EPIMix) is proposed for brain magnetic resonance imaging (MRI). Despite the ultra-fast acquisition speed, EPIMix images suffer from lower signal-to-noise ratio (SNR) and resolution than standard scans.  In this study, we tested whether an unsupervised deep learning framework could improve the image quality of EPIMix exams. We evaluated the proposed network on T2 and T2 FLAIR images and achieved promising qualitative results. The results suggest that deep learning could enable high image quality for ultra-fast EPIMix exams, which could have great clinical utility especially for patients with acute diseases.

4886
Computer 168
Analyzing multi-exponential T2 decay data using a neural network
Hanwen Liu1,2, Roger Tam3,4, John K. Kramer2,5, and Cornelia Laule1,2,4,6

1Physics & Astronomy, University of British Columbia, Vancouver, BC, Canada, 2International Collaboration on Repair Discoveries, Vancouver, BC, Canada, 3Biomedical Engineering, University of British Columbia, Vancouver, BC, Canada, 4Radiology, University of British Columbia, Vancouver, BC, Canada, 5Kinesiology, University of British Columbia, Vancouver, BC, Canada, 6Pathology & Laboratory Medicine, University of British Columbia, Vancouver, BC, Canada

The water molecules within a single voxel may exist in different microenvironments so that the T2 relaxation is considered as a multi-exponential decay. A few quantitative imaging techniques such as myelin water imaging attempt to extract the short T2 component as a marker specific to myelin. However, decomposition of multi-exponential T2 decay data is an ill-posing problem. Commonly used non-negative least squares fitting method is slow, complex and unstable, even with strong regularization and B1 correction. We used synthetic data to train a single neural network for a better and faster analysis of the multi-exponential T2 decay data.

4887
Computer 169
DeepSPIO: A SPIO particles quantification method using Deep Learning
Gabriel della Maggiora1,2,3, Carlos Castillo-Passi1,2,3, Qiu Wenqi4, Masaki Sekino4, Carlos Milovic1,2,3, and Pablo Irarrazaval1,2,3,5

1Department of Electrical Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile, 2Biomedical Imaging Center, Pontificia Universidad Católica de Chile, Santiago, Chile, 3Millennium Nucleus for Cardiovascular Magnetic Resonance, Santiago, Chile, 4Department of Bioengineering, School of Engineering, University of Tokyo, Tokyo, Japan, 5Institute for Biological and Medical Engineering, Pontificia Universidad Católica de Chile, Santiago, Chile

In this study we propose a method to quantify the distribution of Super Paramagnetic Iron Oxide (SPIO) particles with MRI. This task is particularly challenging due to the extreme distortion that these particles produce in the image. Our method is based on a supervised feed-forward deep learning model. The estimation of total quantity of SPIO was in the order of 9% error.  This is potentially useful for detecting breast cancer metastasis by identifying residual particles  in the breast and eventually other organs.

4888
Computer 170
Will a Convolutional Neural Network Trained for Non-contrast Water-Fat Separation Generalize to Post-Contrast Acquisitions?
James W Goldfarb1 and Jie Jane Cao2

1St. Francis Hospital, Roslyn, NY, United States, 2St Francis Hospital, Roslyn, NY, United States

A deep learning CNN trained using precontrast images generalizes to post-contrast images, providing equivalent image quality with fewer swap artifacts.  For wide-spread adoption of deep learning methods, it is important that they have the capability to generalize beyond training data for flexible usage. This work provides important evidence that magnetic resonance deep learning water-fat separation can be used in a variety of settings.  

4889
Computer 171
Fully Automated 3D Body Composition Using Fully Convolutional Neural Networks and DIXON Imaging
Alex Graff1, Dmitry Tkach1, Jian Wu1, Hyun-Kyung Chung1, Natalie Schenker-Ahmed1, David Karow1, and Christine Leon Swisher1

1Human Longevity, Inc, San Diego, CA, United States

Here we show the first fully automated method for body composition profiling with MRI DIXON imaging. The fully automated body composition method developed can be used for radiation-free MRI risk stratification without any manual processing steps making it more accessible clinically. This would be most likely used for risk prediction and risk stratification for diseases such as type II diabetes, cardiovascular disease, and obesity.

4890
Computer 172
Proving b1000 DWI has performance advantage to classify low and high risk Gleason groups by using Neural Network classifier
Hongtao Zhang1, Bo Wang2, Zeyu Hu3, Zhenjie Wu3, Jiamu Xiao3, Gang Wang3, Shulong Wang3, and Huiyi Ye1

1Department of Radiology, Chinese PLA General Hospital, Beijing, China, 2Tsinghua University, Beijing, China, 3Xidian University, Xi'an, China

The Gleason grading of histological samples is recommended for the assessment of prostate cancer risk. Assessing Gleason grade correctly can improve patient prognosis and implement early diagnosis. The aim of this work was to prove that b1000 DWI has the best effect on Gleason high-risk and low-risk grading in T2WI and DWIs with b=1000,b=2000, and b=3000. We use NN (Neural Network) with Ensemble Method on each sequence. The AUC of b1000 DWI was 0.8734, which is significantly higher than those observed for other DWIs.

4891
Computer 173
Harmonization of Longitudinal MRI Scans in the Presence of Scanner Changes
Blake E. Dewey1,2, Can Zhao1, Aaron Carass1, Jiwon Oh3, Peter A Calabresi3, Peter C. M. van Zijl4,5, and Jerry L Prince1,5

1Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, United States, 2Kirby Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, MD, United States, 3Neurology, Johns Hopkins University, Baltimore, MD, United States, 4F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, MD, United States, 5Radiology and Radiological Sciences, Johns Hopkins University, Baltimore, MD, United States

Longitudinal studies are frequently hampered by changes to scanning protocols, forcing research centers to forgo recommended upgrades to scanning equipment, software, and scan protocol design to allow for consistent scanning. Using a harmonization method that utilizes deep learning and a small (n=12) overlap cohort to learn specific differences between structural MR images before and after a significant scanning change and examined longitudinal data acquired annually over 10 years to determine if bias induced by the scanner change is still present after harmonization. We assessed these results using quantitative metrics for contrast and probed volumetric results using automated segmentation algorithms.

4892
Computer 174
Deep Learning with a Novel Surface Feature for Fully Automatic Quantification of Lesion Hyperintensities in Multiple Sclerosis
Peter Adany1, In-Young Choi2,3,4, Scott Belliston3, Jong Chul Ye5, Sharon G. Lynch3, and Phil Lee2,4

1University of Kansas Medical Center, Kansas City, KS, United States, 2Hoglund Brain Imaging Center, University of Kansas Medical Center, Kansas City, KS, United States, 3Neurology, University of Kansas Medical Center, Kansas City, KS, United States, 4Molecular & Integrative Physiology, University of Kansas Medical Center, Kansas City, KS, United States, 5Korea Advanced Institute of Science & Technology, Seoul, Korea, Republic of

Manual lesion segmentation presents major labor and limitations for quantitative MS lesion analysis, and recent improvements in deep learning promise more consistent, fully automatic lesion segmentation. However, convolutional neural networks still rely on learned thresholding of the arbitrary boundaries of diffuse hyperintensities. Therefore, we aimed to develop a new DL framework pairing a CNN and a custom surface feature that could detect hyperintense isocontour in 3 dimensions very sensitively. Our goal is to achieve detection of MS lesions and quantification of lesion hyperintensity volume with a new DL algorithm that combines traditional imaging and a specially designed surface feature.

4893
Computer 175
Tools for assessing and mitigating confounding effects in Machine-Learning based studies of MRI data
Elisa Ferrari1,2,3, Giovanna Spera1, Letizia Palumbo1, and Alessandra Retico1

1INFN, sez. Pisa, Pisa, Italy, 2Scuola Normale Superiore, Pisa, Italy, 3University of Pisa, Pisa, Italy

Using Machine Learning (ML) techniques on neuroanatomical data obtained with magnetic resonance imaging (MRI) is becoming increasingly popular in the study of Psychiatric Disorders (PD). However, this kind of analyses can be affected by overfitting and thus be sensitive to biases in the dataset, producing hardly reproducible results. It is therefore important to identify and correct possible bias sources in the sample. We present two tools aimed at addressing this matter: a methodology to assess the confounding power of a variable in a specific classification task, and a cost function to use during classifier training on highly biased data.


Imaging Myelin+

Exhibition Hall
Thursday 14:45 - 15:45
 Contrast Mechanisms

4894
Computer 1
Inversion Recovery Pointwise Encoding Time Reduction with Radial Acquisition (IR-PETRA) for Direct Myelin Imaging in Human Brain
Hyungseok Jang1, Michael Carl2, Yajun Ma1, Yanjun Chen1, Saeed Jerban1, Eric Y Chang1,3, and Jiang Du1

1Department of Radiology, University of California San Diego, San Diego, CA, United States, 2GE Healthcare, San Diego, CA, United States, 3Radiology Service, VA San Diego Healthcare System, San Diego, CA, United States

Due to very low proton density and rapid signal decay (T2*<300µs at 3T), it is challenging to directly image myelin in the white matter of the brain using MRI. The literature demonstrates that direct myelin imaging is feasible using inversion recovery (IR) preparation followed by dual echo ultrashort echo time (UTE) MRI, allowing direct capture of the rapidly-decaying myelin signal with greatly improved dynamic range.  In this study, we show the efficacy of IR prepared Pointwise Encoding Time Reduction with Radial Acquisition (IR-PETRA) for direct myelin imaging in the human brain.

4895
Computer 2
Myelin UTE imaging, to be or not to be?
Kevin D Harkins1 and Mark D Does1

1Vanderbilt University, Nashville, TN, United States

This work attempts to directly image ultrashort T2 myelin signals using ultra short echo time (UTE) MRI.  Long T2 water signals were suppressed using either adiabatic inversion recovery (AIR) to null signal of a single T1, or multiple adiabatic inversion recovery (MAIR) to null signal over a range of T1s. AIR-UTE showed contrast in white matter, but no such signal was observed in MAIR-UTE. These findings indicate that the AIR-UTE white matter signals are unsuppressed water signals and that the solid proton signals of myelin decay too quickly to be observed by UTE MRI. 

4896
Computer 3
Silent Myelin Imaging with a dipolar-coupled/inhomogeneous MT-Prepared ZTE Radial Sequence
Tobias C Wood1, Emil Ljungberg1, Ana-Beatriz Solana Sanchez2, and Florian Wiesinger1,2

1Neuroimaging, King's College London, London, United Kingdom, 2ASL Europe, GE Healthcare, Munich, Germany

We generated myelin-specific contrast in a silent radial ZTE sequence using a dipolar-coupled MT-prep module. This sequence has great potential for visualising myelin in patient cohorts that do not tolerate the noise from standard MRI, such as infants.

4897
Computer 4
Improved estimates of the g-ratio by modelling its contribution to complex signal evolution in GRE data
Mark Drakesmith1, Elena Kleban1, Fabrizio Fabrizio1,2, and Derek K Jones1

1CUBRIC, Cardiff University, Cardiff, United Kingdom, 2Seimens Healthcare Ltd., Camberley, United Kingdom

g-ratio is an important parameter of axon physiology and there is great interest in estimating it non-invasively in MRI. Existing approaches rely on fitting to a multi-compartment model and calculating g-ratio from the estimated volume fractions (Stikov et al, 2015). Here, we show that we can get improved estimates of the g-ratio by modelling its contribution to frequency offsets in GRE data using a hollow cylinder fibre model. Through simulations and model fitting to in vivo human GRE data we show g-ratio estimates are improved and closer to values obtained from histology compared with the existing approach.

4898
Computer 5
Multi-exponential Relaxometry using $$$\ell_1$$$-regularized Iterative NNLS (MERLIN) for Accurate and Robust Myelin Water Fraction Imaging
Markus Zimmermann1, Ana-Maria Oros-Peusquens1, Zaheer Abbas1,2, Elene Iordanishvili1, Seonyeong Shin1, Seong Dae Yun1, and N. Jon Shah1,2,3,4

1Institute of Neuroscience and Medicine 4, Forschungszentrum Jülich, Jülich, Germany, 2Department of Neurology, RWTH Aachen University, Aachen, Germany, 3Institute of Neuroscience and Medicine 11, JARA, Forschungszentrum Jülich, Jülich, Germany, 4JARA - BRAIN, Translational Medicine, Aachen, Germany

A new parameter estimation algorithm, MERLIN, is presented for accurate and robust multi-exponential relaxometry using MRI. Multi-exponential relaxometry is fundamentally ill-conditioned, and as such, is extremely sensitive to noise. MERLIN is a fully automated, multi-voxel approach that incorporates $$$\ell_1$$$-regularization to enforce sparsity and spatial consistency of the estimated distributions. The proposed method is compared to the conventional $$$\ell_2$$$-regularized NNLS (rNNLS) in simulations and in vivo experiments, using a multi-echo gradient-echo (MEGE) sequence at 3T. The estimated water fraction maps from MERLIN are spatially more consistent, more precise, and more accurate, reducing the root-mean-squared-error by up to 90 percent in simulations.

4899
Computer 6
Feasibility study on artificial neural network based myelin water fraction mapping
Soozy Jung1, Hongpyo Lee1, Kanghyun Ryu1, Jaeeun Song1, Yoonho Nam2, Hojoon Lee3,4, and Donghyun Kim1

1Department of Electrical and Electronic Engineering, Yonsei University, Seoul, Korea, Republic of, 2Seoul St.Mary's Hospital, The Catholic University of korea, Seoul, Korea, Republic of, 3Department of Radiology and Research Institure of Radiological Science, Severance Hospital, Yonsei University College of Medicine, Seoul, Korea, Republic of, 4Department of Radiology, Inje University College of Medicine, Haeundae Paik Hospital, Busan, Korea, Republic of

We developed an artificial neural network (ANN) using magnitude 3-pool signal model based training sets. Simulations were performed for evaluation with various SNR and slice  inhomogeneity (GZ) levels. Two in-vivo data sets were tested. The results show decreased mean error and standard deviation when using the ANN model. The ANN model was more stable than the fitting method for different GZ  values. Moreover, the processing time of the ANN model took 140 times less than the fitting method.

4900
Computer 7
Sensitivity of in vivo myelin imaging techniques to detect subtle changes in myelin lipid and protein content in post-mortem multiple sclerosis brain tissues
Vanessa Wiggermann1,2,3, Verena Endmayr4, Enedino Hernandez-Torres2,3, Romana Höftberger5, Gregor Kasprian6, Alexander Rauscher1,2,3,7, and Simon Hametner8

1Physics and Astronomy, University of British Columbia, Vancouver, BC, Canada, 2Pediatrics, University of British Columbia, Vancouver, BC, Canada, 3UBC MRI Research Centre, Univeristy of British Columbia, Vancouver, BC, Canada, 4Neuroimmunology, Medical University of Vienna, Vienna, Austria, 5Institute of Neurology, Medical University of Vienna, Vienna, Austria, 6Biomedical Imaging and Image‐Guided Therapy, Medical University of Vienna, Vienna, Austria, 7Radiology, Univeristy of British Columbia, Vancouver, BC, Canada, 8Neuropathology, Medical University of Göttingen, Göttingen, Germany

Previous post-mortem single-slice myelin water fraction (MWF) measurements have shown good correlations with the myelin lipid fraction across tissue types. However, the role of protein content was not assessed nor have validations been performed for the whole brain 3D-Gradient and Spin Echo (GraSE) technique that has been employed in recent studies. We showed that 3D-GraSE based MWF measurements reliably distinguished regions of different myelin integrity reflective of difference in myelin lipid and protein content. In contrast, subtle variations in MWF within tissue classes or between persons may relate to differences in protein content.

4901
Computer 8
Robust myelin water imaging from multi-echo T2 data using second-order Tikhonov regularization with control points
Erick Jorge Canales-Rodriguez1,2, Marco Pizzolato3, Gian Franco Piredda1,3,4, Tom Hilbert1,3,4, Kunz Nicolas5, Tobias Kober1,3,4, Jean-Philippe Thiran1,3, Caroline Pot6,7, and Alessandro Daducci1,3,8

1Department of Radiology, Centre Hospitalier Universitaire Vaudois (CHUV), Lausanne, Switzerland, 2FIDMAG Germanes Hospitalàries Research Foundation, Barcelona, Spain, 3LTS5, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, 4Advanced Clinical Imaging Technology, Siemens Healthcare AG, Lausanne, Switzerland, 5Centre d'Imagerie BioMédicale (CIBM)-AIT, École polytechnique fédérale de Lausanne (EPFL), Lausanne, Switzerland, 6Department of Pathology and Immunology, Geneva University Hospital and University of Geneva, Geneva, Switzerland, 7Division of Neurology and Neuroscience Research Center, Department of Clinical Neurosciences, Centre Hospitalier Universitaire Vaudois (CHUV), Lausanne, Switzerland, 8Computer Science Department, University of Verona, Verona, Italy

Myelin water imaging is an MRI technique used to quantify myelination in the brain. The state-of-the-art reconstruction method is based on non-negative least squares optimization with zero-order Tikhonov regularization. In this study, a second-order Tikhonov regularization approach with control points was examined. This penalty term is more efficient for promoting smooth solutions while minimizing the contamination between myelin and non-myelin components. The performance of the proposed algorithm was investigated on in-vivo and ex-vivo multi-echo T2 data. It exhibited a higher correlation with histology than the state-of-the-art method. Its stability was studied using scan-rescan data.

4902
Computer 9
A Model-Based Method for Estimation of Myelin Water Fractions
Yudu Li1,2, Rong Guo1,2, Yibo Zhao1,2, Yang Chen1,3, Bryan Clifford1,2, Tianyao Wang4, Chenyan Wang1,5, Yiping Du6, Yao Li6,7, and Zhi-Pei Liang1,2

1Beckman Institute for Advanced Science and Technology, University of Illinois at Urbana-Champaign, Urbana, IL, United States, 2Department of Electrical and Computer Engineering, University of Illinois at Urbana-Champaign, Urbana, IL, United States, 3Department of Mathematics & Statistics, Xi’an Jiaotong University, Xi'an, China, 4Department of Radiology, The Fifth People's Hospital of Shanghai, Shanghai, China, 5Institute for Medical Imaging Technology (IMIT), School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China, 6School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China, 7Med-X Research Institute, Shanghai Jiao Tong University, Shanghai, China

Quantitative mapping of myelin water fractions (MWF) can substantially improve our understanding of the progression of several demyelination white matter diseases such as multiple sclerosis. While MWF can be determined from both T2-weighted and T2*-weighted imaging data, it is much faster to collect T2*-weighted imaging data. However, estimation of MWF from T2*-weighted imaging data using a multi-exponential component model is an ill-conditioned problem whose solutions are often very sensitive to noise and modeling errors. In this work, we address this problem using a new model-based method. This method is characterized by: a) absorbing the spectral priors using the Bayesian-based statistical framework, and b) absorbing the spatial priors using a reproducing kernel based model. Both simulation and experimental results show the proposed method significantly outperforms the conventional parameter estimation methods currently used for MWF estimation.

4903
Computer 10
Analysis of Gradient Echo Myelin Water Imaging (GRE-MWI) for water exchange and scan parameters
Hyeong-Geol Shin1, Se-Hong Oh2, Joon Yul Choi1, Kyeongseon Min1, Hyunsung Eun1, and Jongho Lee1

1Department of Electrical and Computer Engineering, Seoul National University, Seoul, Korea, Republic of, 2Biomedical Engineering, Hankuk University of Foreign Studies, Yongin, Korea, Republic of

In this study, we investigate the effects of the compartmental water exchange on gradient echo myelin water imaging (GRE-MWI). We simulate MWF variation from different scan parameters (flip angle and TR) using a four pool white matter model and compare the simulation results with the in-vivo measurements. The results demonstrate that 1) the simulation with the water exchange better explains the in-vivo results and 2) GRE-MWI with a long TR can provide robust myelin water quantification regardless of changes in flip angle. Therefore, our results suggest GRE-MWI with a long TR as a robust option for myelin water imaging. 

4904
Computer 11
Influence of model settings on myelin water fraction and frequency distribution for gradient-echo MRI at 7 Telsa
Kiran Thapaliya1, Viktor Vegh1, Steffen Bollmann1, and Markus Barth1

1The University of Queensland, Brisbane, Australia

Quantitative assessment of model parameters (water fraction and frequency shift) estimated using a multi-compartment model can be useful to study tissue properties in white matter. In this work, we utilise multi-compartment models for multi-echo gradient echo data acquired at 7T. We investigate the variation of model parameters that could potentially be affected by differences in tissue microstructure in the corpus callosum. We further study the effect of different models (number of compartments and parameters) on the estimation of tissue parameters. We show that the tissue parameters vary across the sub-regions of the corpus callosum and are effected by different modelling choices.

4905
Computer 12
Gradient echo modelling with macroscopic field variations and large flip angles
Martin Soellradl1, Stefan Ropele1, and Christian Langkammer1

1Department of Neurology, Medical University of Graz, Graz, Austria

The signal decay of a 2D gradient echo sequence is substantially influenced by macroscopic field variations along the slice profile. Here we propose a numerical model describing the signal decay due to a macroscopic field gradient for arbitrary excitation pulses with large flip angles. Phantom and in-vivo experiments show that accurate modelling requires inclusion of the phase along the slice profile and the polarity of the slice selection gradient. Additionally, we show that applying the model yields better results for R2*-mapping and myelin water fraction estimation.

4906
Computer 13
In vivo assessment of the anisotropy of R2* maps in white matter
Renat Sibgatulin1, Andreas Deistung1, Daniel Güllmar1, Christoph Birkl2, Stefan Ropele3, and Jürgen Rainer Reichenbach1,4,5,6

1Institute of Diagnostic and Interventional Radiology, Jena University Hospital – Friedrich Schiller University Jena, Jena, Germany, 22 University of British Columbia, Vancouver, BC, Canada, 3Department of Neurology, Medical University of Graz, Graz, Austria, 4Michael Stifel Center for Data-driven and Simulation Science Jena, Friedrich Schiller University Jena, Jena, Germany, 5Abbe School of Photonics, Friedrich Schiller University Jena, Jena, Germany, 6Center of Medical Optics and Photonics, Friedrich Schiller University Jena, Jena, Germany

The effective transverse relaxation rate (R2*) is increasingly used in quantitative MRI, and its dependence on the orientation of white matter fibers in the brain has received significant attention. In this contribution, we assess the effect of the flip angle of a multi-echo gradient-echo sequence on the orientation dependence of the derived R2* map and suggest a simplified explanation to the observed R2*(θ; FA) behavior.

4907
Computer 14
Estimation of microstructural properties of white matter from multiple orientation GRE signal simulations of realistic models
Renaud Hedouin1, Kwok-Shing Chan1, Riccardo Metere1, and Jose P Marques1

1Donders institute, Radboud university, Nijmegen, Netherlands

This study presents the creation of 2D white matter models, based on real histologically derived axon shapes, with large range of microstructure parameters (FVF, g-ratio). These models are used to simulate the complex gradient echo signal evolution under different main magnetic field orientations for (amongst other parameters) varying magnetic susceptibility and water density in the myelin compartment. A deep learning network, trained from those data, shows its capacity to recover parameter microstructure properties as g-factor and susceptibility on test data.

4908
Computer 15
Evaluating the sensitivity of T1w/T2w, MTR, MWF and DKI to variation of myelin content
Run Pu1, Hongjian He1, Zhe Wu2, Yi Sun3, and Jianhui Zhong1,4

1Center for Brain Imaging Science and Technology, Key Laboratory for Biomedical Engineering of Ministry of Education, College of Biomedical Engineering and Instrumental Science, Zhejiang University, HangZhou, China, 2Department of Neurology and Neurosurgery, McConnell Brain Imaging Centre, Montreal Neurological Institute and Hospital, McGill University, Montreal, QC, Canada, 3MR Collaboration, Siemens Healthcare Ltd., Shanghai, China, 4Department of Imaging Sciences, University of Rochester, Rochester, NY, United States

MRI metrics such as T1w/T2w ratio, magnetization transfer ratio (MTR), myelin water faction (MWF) and diffusion kurtosis imaging (DKI) indices have been used to detect myelin content. To assess the sensitivity of above metrics to variation of myelin content, in vivo human corpus callosum is used as a test case in the study. The results suggest that MTR varies least but MWF varies the most as myelin content changes.

4909
Computer 16
Stain-free histology to validate quantitative MRI
Gabriel Mangeat1, Harris Nami1, Nicolas Pinon1, Alexandru Foias1, Nikola Stikov1,2, Tobias Granberg3,4, and Julien Cohen-Adad1,5

1Ecole Polytechnique de Montreal, Montreal, QC, Canada, 2Montreal Health Institute, Montreal, QC, Canada, 3Department of Clinical Neuroscience, Karolinska Institutet, Sweden, Sweden, 4Division of Neuroradiology, Department of Radiology, Karolinska University Hospital, Stockholm, Sweden, 5Functional Neuroimaging Unit, CRIUGM, Université de Montréal, Montreal, QC, Canada

Quantitative MRI (qMRI) is reproducible but often lacks calibration and/or specificity to the underlying microstructure. Light transmission optical histology of stained tissue is a popular method for validation, however, it is hampered by calibration issues and inhomogeneous penetration of staining agents. We propose a method to validate quantitative MRI metrics using stainless histology by utilizing the innate autofluorescence spectra of tissues when excited with ultraviolet laser. We demonstrate a proof-of-concept application of a qMRI validation pipeline on a pig spinal cord section with in vivo and ex vivo qMRI followed by histological autofluorescence microscopy to quantify myelin content.

4910
Computer 17
Apparent Population Inversion Due to Steady-State Transcytolemmal Water Exchange
Xin Li1, Silvia Mangia2, Jing-Huei Lee3, Ruiliang Bai4, and Charles S. Springer1

1Advanced Imaging Research Center, Oregon Health & Science University, Portland, OR, United States, 2Center for Magnetic Resonance Research, University of Minnesota, Minneapolis, MN, United States, 3Biomedical Engineering, University of Cincinnati, Cincinnati, OH, United States, 4Interdisciplinary Institute of Neuroscience and Technology, Zhejiang University, Hangzhou, China

The homeostatic cellular water efflux rate constant, kio, has a significant contribution from cell membrane sodium pump activity previously unmeasurable.  With high extracellular contrast agent concentration or ultra-low magnetic field, kio can be precisely determined by two-site-exchange analysis of in vivo 1H2O longitudinal relaxation data.  With the low field case, there is an inversion of the apparent tissue compartmental contributions from the true values.  The NMR shutter-speed organizing principle informs an analysis spanning the entire range of conditions.  

4911
Computer 18
An inhomogeneous magnetization transfer (ihMT) quantification method robust to B1 and T1 variations in magnetization prepared acquisitions
Gopal Varma1, Fanny Munsch1, Olivier M Girard2, Guillaume Duhamel2, and David C Alsop1

1Radiology, Division of MR Research, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, United States, 2Aix Marseille Univ, CNRS, CRMBM, Marseille, France

Standard MT and ihMT ratio (ihMTR) measures can be sensitive to Band T1, making them less specific to tissue microstructure. Using the inverse signal, i.e. one divided by the signal, and a high flip-angle reference image in calculation of an ihMTR metric has been proposed as a metric with improved insensitivity to T1 and B1 in steady-state gradient-echo sequences. We present a modified method for use in prepared sequences such as magnetization prepared rapid gradient echo (MPRAGE). The sensitivity of ihMT MPRAGE metrics to T1 and B1 was tested using simulations and acquisitions in brains of healthy volunteers.

4912
Computer 19
In vivo inhomogeneous magnetization transfer (ihMT) outside the brain using radial ultra-short echo-time acquisitions
Gopal Varma1, Cody Callahan1, Olivier M Girard2, Guillaume Duhamel2, Aaron K Grant1, and David C Alsop1

1Radiology, Division of MR Research, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, United States, 2Aix Marseille Univ, CNRS, CRMBM, Marseille, France

Inhomogeneous magnetization transfer (ihMT) effects have been readily observed in myelinated structures. The advent of low duty-cycle ihMT to increase the signal allows application of ihMT in other tissues. In this work, we explore the feasibility of applying ihMT in non-myelinated tissues such as the heart, liver, and kidneys of mice. This is achieved using a radial, ultra-short echo-time acquisition for greater motion robustness. The results demonstrate a measurable ihMT signal outside the central nervous system. Thus the microstructure of such tissues might be assessed based on the dipolar order contribution to ihMT.

4913
Computer 20
Myelin-sensitive imaging of the optic chiasm and optic nerve at 3T using inhomogeneous magnetization transfer (ihMT) with high B1 pulses
Ece Ercan1, Fang Yu1, Ivan E. Dimitrov2,3, Gopal Varma4, David C. Alsop4, Robert E. Lenkinski1,2, and Elena Vinogradov1,2

1Department of Radiology, UT Southwestern Medical Center, Dallas, TX, United States, 2Advanced Imaging Research Center, UT Southwestern Medical Center, Dallas, TX, United States, 3Philips Healthcare, Gainesville, FL, United States, 4Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, United States

Inhomogeneous magnetization transfer (ihMT) imaging is a novel enhanced magnetization transfer contrast, which has been shown to originate from long-lived dipolar couplings in the tissue (e.g. dipolar couplings between the methylene molecules of the myelin phospholipid bilayer). In this study, we optimized an ihMT scan protocol for imaging the optic nerve and chiasm for the first time. This method may potentially be used for quantitative evaluation of patients with multiple sclerosis (MS), as well as other diseases affecting the visual pathway.

4914
Computer 21
Investigating the Influence of Adipose Fat on the Inhomogeneous Magnetization Transfer (ihMT) Images
Ece Ercan1, Gopal Varma2, Ivan E. Dimitrov3,4, Marco C. Pinho1,3, Shu Zhang1, Xinzeng Wang1, Ananth Madhuranthakam1,3, David C. Alsop2, Robert E. Lenkinski1,3, and Elena Vinogradov1,3

1Department of Radiology, UT Southwestern Medical Center, Dallas, TX, United States, 2Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, United States, 3Advanced Imaging Research Center, UT Southwestern Medical Center, Dallas, TX, United States, 4Philips Healthcare, Gainesville, FL, United States

Inhomogeneous Magnetization Transfer (ihMT) imaging is a novel enhanced magnetization transfer technique. In this study, we investigated the influence of fat (i.e. adipose tissue) and echo time on the ihMT ratio through simulation, phantom, and in vivo studies.  A substantial variation on the ihMTR values in the presence of fat is illustrated, depending on the echo times used.

4915
Computer 22
Assessment of two T1D components within myelinated tissue with ihMT MRI
Victor N. D. Carvalho1,2, Olivier M. Girard1, Andreea Hertanu1, Samira Mchinda1, Lucas Soustelle1, Axelle Grélard3, Antoine Loquet3, Erick J. Dufourc3, Gopal Varma4, David C. Alsop4, Pierre Thureau2, and Guillaume Duhamel1

1Aix Marseille Univ, CNRS, CRMBM UMR 7339, Marseille, France, 2Aix Marseille Univ, CNRS, ICR UMR 7273, Marseille, France, 3CBMN UMR 5248, CNRS University of Bordeaux, Pessac, France, 4Department of Radiology, Division of MR Research, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, United States

T1D, the relaxation time of the dipolar order, is a probe for membrane dynamics and organization that could be used to assess myelin integrity. A single-component T1D model associated with a modified ihMT sequence had been proposed for in vivo evaluation of T1D with MRI. However, experiments and simulations revealed that myelinated tissues exhibit multiple T1D components. A bi-component T1D model is proposed and validated. Fits in a rat spinal cord yield two T1Ds of about 10 ms and 400 μs. The results suggest that myelin has a dynamically heterogeneous organization.

4916
Computer 23
Reproducibility of inhomogeneous magnetization transfer (ihMT): a test-retest, multi-site study
Lei Zhang1, Huipeng Ren1, Qing Fan1, Xiaocheng Wei2, and Zhuanqin Ren1

1Baoji Center Hospital, Baoji, China, 2GE Healthcare China, Beijing, China

Derived from conventional magnetization transfer, inhomogeneous magnetization transfer (ihMT) has been shown to be a promising method for myelin imaging in recent studies. In the present study, the test-retest reproducibility and multi-site variability of ihMT in measuring major white matter fibers were evaluated. Good test-retest reproducibility and multi-site agreements were obtained. These findings support the use of ihMT measurements as biomarkers in multicenter and/or longitudinal studies.

4917
Computer 24
On the dipolar order underlying broad macromolecular lines
Olivier M. Girard1, Victor N.D. Carvalho1, Ludovic de Rochefort1, Andreea Hertanu1, Pierre Thureau2, Gopal Varma3, David Alsop3, and Guillaume Duhamel1

1Aix Marseille Univ, CNRS, CRMBM, Marseille, France, 2Aix Marseille Univ, CNRS, ICR, Marseille, France, 3Division of MR Research, Beth Israel Deaconess Medical Center, Harvard Medical School, Boston, MA, United States

Dipolar order has recently regained attention in MRI to analyze dipolar broadened lines in CEST and inhomogeneous Magnetization Transfer (ihMT), leading to new frequency irradiation patterns for enhanced saturation and access to an unexplored degree of freedom. A better understanding of dipolar order is of great interest to guide intuition and may lead to fundamental optimization of the ihMT technique, which is a promising tool providing new tissue contrasts. In this contribution we propose to review this concept, considering a simplified model of isolated proton pairs and the general Provotorov theory of RF saturation which applies to an ensemble of coupled spin.

4918
Computer 25
Non-invasive detection of molecular profiles in the human brain.
Shir Filo1, Oshrat Shtangel1, and Aviv A Mezer1

1The Edmond and Lily Safra Center for Brain Sciences, The Hebrew University of Jerusalem, Jerusalem, Israel

Lipids makes more than 40% of the human brain in dry weight, and have broad information carrying roles in the CNS. In-vivo quantitative MRI (qMRI) aims at characterizing the biological properties of brain tissue. However,