RM-GROWL-GRASP: Image Registration Involved Two-step Motion Compensation System for Real-time Non-Cartesian Liver DCE-MRI
Zhifeng Chen1,2, Peiwei Yi1,2, Zhongbiao Xu1,2, Jucheng Zhang3, Yingjie Mei1,2, Xia Kong4, Zhenguo Yuan5, Yaohui Wang6, Ling Xia7, Yanqiu Feng1,2, and Feng Liu8

1School of Biomedical Engineering, Guangdong Provincial Key Laboratory of Medical Image Processing, Southern Medical University, Guangzhou, China, 2Key Laboratory of Mental Health of the Ministry of Education, Southern Medical University, Guangzhou, China, 3Department of Radiology, 2nd Affiliated Hospital, School of Medicine, Zhejiang University, Hangzhou, China, 4School of Computer and Information Science, Hubei Engineering University, Wuhan, China, 5Shandong Medical Imaging Research Institute, Shandong University, Jinan, China, 6Division of Superconducting Magnet Science and Technology, Institute of Electrical Engineering, Chinese Academy of Sciences, Beijing, China, 7Department of Biomedical Engineering, Zhejiang University, Hangzhou, China, 8School of Information Technology and Electrical Engineering, The University of Queensland, Brisbane, QLD, Australia


Motion is an inescapable problem in abdominal MRI. Involuntary organ movements caused mainly by respiratory often results in motion artifacts and image details blurring in liver MRI. For dynamic imaging, motion also harms temporal information. Recently, high spatiotemporal resolution free-breathing liver DCE-MRI have attracted much attentions of radiologists and scholars. We propose to combine mutual-information-based image registration with motion-sorted GROWL-GRASP approach for golden-angle radial liver DCE-MRI, which enable free-breathing imaging. The results demonstrate that better image quality including SNR benefit, lower motion artifacts and more diagnostic information can be generated compared to current motion compensation methods.

Target Audience

Radiologist and scholars who are interested in real-time DCE-MRI and motion-corrected abdominal imaging


Motion is a challenging problem in abdominal MRI1, 2. Motion artifacts influence the imaging result in a great manner, sometimes can lead to misdiagnosis. DCE-MRI is a powerful tool to detect the liver diseases like liver cirrhosis and hepatocellular carcinoma. Previous studies partly solve the abdominal motion issues through golden-angle radial sampling in liver DCE-MRI3, 4. However, it still suffers from remaining motion artifacts and image quality need to be improved. In this study, we propose to involve image registration approach5 into motion-sorted GROWL-based method6, for a further enhancement of motion-corrected abdominal DCE-MRI. Two radiologists proved that the proposed scheme resulted in significantly better image quality and diagnostic information compared to those tested motion correction approaches.


In this study, the main objective function is expressed as following:$$I=argmin||G-1(FSI)-m||22 +λ||C(I)||1+μ||M(I)||1 [X1]$$

Where G is the GROWL operator6, F is the non-uniform FFT operator, S represents coil sensitivity map, m is the measured dataset, C and M represent the sparsity transform along dynamic contrast-enhanced dimension (t-dimension) and motion dimension, respectively. λ and μ are regularization parameters. I is the to-be-restored image set. The overall structure of framework is shown as Fig.1, and it consists of the following steps: (i) the radial k-space raw data is transformed into propeller trajectory using GROWL operator6. (ii) motion-state signal is computed from central points of each spoke, then the previous propeller data is sorted according to the motion states4. (iii) the density correction function is calculated based on the new trajectory7. (iv) all the known conditions are embedded into Eqn.[X1]. (v) mutual-information-based registration operation5 is performed motion-state-by-motion-state for all motion states through exploiting DCE-averaged prior image.

An in vivo liver DCE imaging experiment was performed on a 3.0 T Skyra MR scanner (Siemens AG, Erlangen, Germany) using a body/spine coil array with 20 elements. A radial stack-of-stars 3D FLASH pulse sequence with free-breathing golden-angle scheme was applied for this acquisition. The relevant parameters included: FOV 380×380 mm2, TR/TE = 3.78/1.73 ms, partitions 42, number of readout points in each spoke 512, oversampling ratio 2, number of spokes 2108, slice thickness 3 mm, and flip angle 12°. Frequency-selective fat suppression was used for this acquisition. Gradient-delay errors were corrected before image reconstruction8.

In this work, the selected slice (Slice 17) of the dataset was sorted into 4 motion states, 15 DCE phases, 34 spokes/frame for all methods. All image reconstructions were implemented in Matlab (R2014a; the Mathworks, Natick, MA, USA), for off-line reconstruction on an HP workstation (Z820) with 12-Core 2.10-GHz Intel Xeon E5-2620 v2 CPU and 128GB of Memory. For image quality evaluation, two experienced radiologists were blinded to score all the images. The ANOVA approach was used to test the differences among all reconstructions using Excel.


As shown in Fig.2, the reconstructed results of NUFFT, iGRASP, XD-GRASP and the proposed RM-GROWL-GRASP are all displayed. The light blue lines show the movements of liver among different motion states. The streaking artifacts are obviously decreased in the proposed scheme (see blue boxes). Image details and edge boundary are better as the red arrows pointed. The scores of each method are listed in Fig.3, together with statistical results. The image quality of the proposed approach is significantly better than other methods. From the image quality comparison, it is demonstrated that RM-GROWL-GRASP>XD-GRASP>iGRASP>NUFFT (P<0.05).


As we know, non-Cartesian trajectory like golden-angle radial sampling is insensitive to motion and can reduce motion artifacts to some extent. However, it can distribute motion to all directions, leading to motion-averaged effects and image details blurring 4, 9. The proposed scheme uses a two-step motion compensation scheme, that brings more accurate motion estimation than conventional motion correction methods. Moreover, it can be seen in Fig.2 that RM-GROWL-GRASP scheme leads to higher SNR than iGRASP under the same condition, because more data are used in image reconstruction procedure (propeller data vs. radial data). In addition, the registration operator also suppresses the noise and make the background clean (Fig.2).


The proposed technique is a two-step motion compensation scheme, which is effective for free-breathing abdominal DCE-MRI. Compared to original golden-angle radial technique, it offers higher image quality, that is, less motion artifacts, clearer image detail, and more anatomical information. Unlike one-step motion-sorting techniques, it has better SNR benefit and performs more accurate motion correction. The proposed scheme is particularly suitable for irregular respiratory movements and as a good supplement to motion-information-sorted abdominal DCE-MRI, it can be of great potential value for real-time liver DCE-MRI clinical application.


We thank Dr. Li Feng in NYU for his constructive suggestion about density compensation function suggestion and motion estimation. This study was supported by NSFC grant (61801205) and CPSF grant (2018M633073).


[1] Wood ML, Henkelman RM. MR image artifacts from periodic motion. Med Phys. 1985;12(2):143–151.

[2] Lin W, Guo J, Rosen MA, et al. Respiratory Motion-Compensated Radial Dynamic Contrast-Enhanced (DCE)-MRI of Chest and Abdominal Lesions. Magn Reson Med. 2008; 60(5): 1135–1146.

[3] Feng L, Grimm R, Block KT, et al. Golden-angle radial sparse parallel MRI: combination of compressed sensing, parallel imaging, and golden-angle radial sampling for fast and flexible dynamic volumetric MRI. Magn Reson Med. 2014; 72(3): 707–717.

[4] Feng L, Axel L, Chandarana H, et al. XD-GRASP: Golden-angle radial MRI with reconstruction of extra motion-state dimensions using compressed sensing. Magn Reson Med. 2016; 75(2): 775–788.

[5] Maes F, Collignon A, Vandermeulen D, et al. Multimodality image registration by maximization of mutual information. IEEE Trans Med Imaging. 1997; 16(2): 187–198.

[6] Lin W, Huang F, Li Y, et al. GRAPPA operator for wider radial bands (GROWL) with optimally regularized self-calibration. Magn Reson Med. 2010; 64(3): 757–766.

[7] Pipe JG, Menon P. Sampling density compensation in MRI: rationale and an iterative numerical solution. Magn Reson Med. 1999; 41(1): 179–186.

[8] Block KT, Uecker M. Simple method for adaptive gradient-delay compensation in radial MRI. In Proceedings of the 19th Annual Meeting of ISMRM, Montreal, Canada, 2011. p. 2816.

[9] Glover GH, Pauly JM. Projection reconstruction techniques for reduction of motion effects in MRI. Magn Reson Med. 1992;28(2):275–289.


FIG.1. The flowchart of the proposed scheme. The GROWL workflow is displayed in the dotted box.

FIG.2. Comparison of NUFFT, iGRASP, XD-GRASP and proposed RM-GROWL-GRASP methods for the same in vivo liver DCE data. Motion Sn represents the motion state.

FIG.3. Comparison of the image quality of NUFFT, iGRASP, XD-GRASP and proposed RM-GROWL-GRASP methods, P < 0.05 is deemed to be significantly difference.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)