Reducing g-factor for TGRAPPA accelerated real-time cardiac cine imaging
Sen Jia1,2, Haifeng Wang1, Xin Liu1, Hairong Zheng1, and Dong Liang1,2

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


TGRAPPA acceleration alleviates the intense tradeoff between spatial and temporal resolutions for real-time cardiac cine imaging. However, it suffers from significant noise amplification due to ill-conditioned inverse reconstruction at high acceleration factors. A quadruple extended TGRAPPA reconstruction model is established to jointly utilize the additional spatial encoding capability of background phase and the high-order noise model by nonlinear kernel method. Prospective real-time cine experiments showed superior noise suppression of this non-iterative technique at 6-8X acceleration.


TGRAPPA is the commercially available acceleration method to alleviate the intense tradeoff between spatial and temporal resolutions for real-time cardiac cine imaging1. However, it suffers from significant noise amplification due to ill-conditioned inverse reconstruction at high acceleration factors. A non-iterative reconstruction technique named as CarRace (Cardiac Cine using conjugated and nonlinear virtual coils) is proposed to significantly improve the inverse condition for real-time cardiac cine with high temporal resolutions.


TGRAPPA reconstruction includes two steps: (1) GRAPPA kernel calibration using the ACS data generated by averaging all time-interleaved time frames; (2) frame by frame data synthesis by convolving the GRAPPA kernel with acquired data. CarRace proposes to transform the virtual conjugate coil (VCC) concept extended dataset 2,3 to a high-dimensional feature space by nonlinear kernel (NLK) method 4. A quadruple extended GRAPPA reconstruction model is then established as shown in Figure 1. Inverse condition is jointly improved by utilizing the additional spatial encoding capability of background phase and by suppressing the nonlinear bias from noise in the time averaged ACS data.


In-vivo experiments were IRB approved with written informed consents obtained from all imaged subjects. Prospective free-breathing, real-time cardiac cine imaging studies were performed on 3T UIH uMR 770 scanner (United Imaging, Shanghai, China) with a 28-channel cardiac coil. Balanced steady-state free precession (BSSFP) sequence with a time-interleaved phase encoding scheme was used for real-time data acquisition. Two healthy volunteers were prospectively recruited and experimented with 4-8-fold acceleration factors in short-axis and four-chamber views. The imaging parameters were TE/TR = 1.23/2.76 ms, FOV = 330×250 mm2 , slice thickness = 8 mm, 50 cardiac phases, base resolution = 192, and phase resolution = 75%. The average of all frames along the temporal direction severed as the ACS data. Temporal resolution was improved from 86.8 ms per frame at 4X acceleration to 43.4 ms per frame at 8X acceleration. All accelerated datasets were reconstructed offline by traditional TGRAPPA and proposed CarRace methods respectively. These two methods used the same kernel size: 3 (phase encoding) x 9 (readout). Geometry factor maps were calculated to quantify their noise performance in the reconstructed images using the pseudo replica method with 300 replicas 5.


Multiple GRAPPA kernels are required by CarRace for different time frames due to the shifted sampling positions in virtual conjugate coil, relative to the real sampling positions in physical coil as shown in Figure 2. CarRace reduces the mean g-factors by nearly 39% and achieved nearly homogeneous noise distribution across both the spatial and time dimensions. CarRace enables flexible control of the number of virtual coils used for reconstruction and gives NL-TGAPPA and VCC-TGRAPPA methods separated when only nonlinear or conjugate virtual coils are used. Figure 3 illustrates the comparison of reconstruction quality between these methods. CarRace achieves the best visual image quality. The reconstruction quality by CarRace is compared with TGRAPPA at different acceleration factors in Figure 4. The noise in both image space and temporal profile are well suppressed by CarRace and benefited the depiction of image and temporal details at 6X and 8X accelerations.


Data redundancy originated from background phase and nonlinear mapping were jointly integrated with TGRAPPA to give a new CarRace reconstruction technique for real-time cardiac cine imaging. Inverse condition at high acceleration factors was improved, and noise was well suppressed in the final dynamic images. The proposed CarRace utilized the virtual coil implementation to enable a noniterative solution to the inverse problem with comparable computational burden as traditional TGRAPPA.


This work was partially supported by the National Natural Science Foundation of China (61471350, 81729003), the Natural Science Foundation of Guangdong (2018A0303130132) and the Basic Research Program of Shenzhen (JCYJ20150831154213680).


  1. Breuer FA, Kellman P, Griswold MA, Jakob PM. Dynamic Autocalibrated Parallel Imaging using Temporal GRAPPA (TGRAPPA). Magn Reson Med 2005; 53:981-985.
  2. Blaimer M, Heim M, Neumann D, Jakob PM, Kannengiesser SA, Breuer FA. Comparison of Phase-Constrained Parallel MRI Approaches: Analogies and Differences. Magn Reson Med 2016; 75:1086-1099.
  3. Blaimer M, Gutberlet M, Kellman P, Breuer FA, Kostler H, Griswold MA. Virtual Coil Concept for Improved Parallel MRI Employing Conjugate Symmetric Signals. Magn Reson Med 2009; 61:93-102.
  4. Chang Y, Liang D, Ying L. Nonlinear GRAPPA: A Kernel Approach to Parallel MRI Reconstruction. Magn Reson Med 2012; 68:730-740.
  5. Robson PM, Grant AK, Madhuranthakam AJ, Lattanzi R, Sodickson DK, McKenzie CA. Comprehensive Quantification of Signal-to-Noise Ratio and g-Factor for Image-Based and k-Space-Based Parallel Imaging Reconstructions. Magn Reson Med 2008; 60:895-907.


Figure 1. The schematic description of proposed CarRace algorithm for improved dynamic parallel imaging reconstruction. (a) Generating nonlinear kernel mapped and phase conjugated autocalibration (ACS) data in virtual coils. (b) Constructing a convolution kernel based on extended ACS data. (c) Data synthesis formula using both physical and virtual coils for reconstructing missing samples in physical coil.

Figure 2. The g-factor maps for adjacent three frames of 6X accelerated real-time cine reconstructed by CarRace and TGRAPPA. Multiple kernel calibrations (depends on the acceleration factors) are needed for CarRace due to the shifted positions of known data between physical coil and conjugate virtual coil. The g-factor maps calculated by pseudo replica method varied slightly between adjacent frames indicates that CarRace achieves homogeneous noise behavior across the time dimension.

Figure 3. Comparing the reconstruction quality between traditional TGRAPPA, NL-, VCC-extended TGRAPPA and proposed CarRace methods on a 8X accelerated time-interleaved real-time cine dataset. The proposed virtual coils implementation of CarRace algorithm enables flexible switch between these algorithms. CarRace achieves the best visual reconstruction quality among these methods.

Figure 4. Prospectively comparing proposed CarRace with traditional TGRAPPA reconstruction at 4X, 6X and 8X acceleration rates for time-interleaved real-time cine. The imaging parameters: TR/TE = 2.75/1.25 ms, acquisition matrix Size = 192 x 124, FOV = 330 x 330 mm, 28 channel cardiac coil on a 3T scanner.

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