4745

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

Synopsis

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.

Introduction

Real-time, phase-contrast (rt-PC) MRI has the potential to overcome the limitations, such as sensitivity to arrhythmia and patient motion, associated with standard ECG-gated PC MRI.1,2 Conventional rt-PC MRI, however, produces relatively low spatial resolution1,3 and/or temporal resolution.3,4 making it unreliable, especially for imaging pediatric patients with smaller hearts and faster heart rates. One approach to address these limitations is to highly accelerate rt-PC MRI using compressed sensing (CS)5 and radial k-space sampling.6 Unfortunately, a standard iterative reconstruction of highly-accelerated (e.g. acceleration factor [R] >30) radial k-space data using a framework like Golden-angle radial sparse parallel (GRASP)7 is likely to produce blurring and require a lengthy reconstruction time, owing to a need to grid and regrid data between radial and Cartesian k-space for each iteration. One approach to accelerate the reconstruction speed is to grid the radial data onto a Cartesian space once using a GRAPPA operator, such as in GROG-GRASP.8 The performance of GRASP and GROG-GRASP for highly-accelerated (R> 30) data have not been reported. In this study, we implemented a variant of GROG-GRASP that uses NUFFT instead, in order to achieve a good balance between data fidelity and reconstruction speed.9 The purpose of this study was to develop the proposed Cartesian GRASP framework and compare its performance to GRASP and GROG-GRASP frameworks for reconstructing highly-accelerated rt-PC MRI data obtained from pediatric patients.

Method

Pulse Sequence: We evaluated our proposed rt-PC acquisition and reconstruction methods in 7 pediatric patients (3 boys, mean age = 9.4 ± 2.6 years) undergoing a clinical cardiac MRI at 1.5T scanner (Siemens, Aera), which included standard PC MRI at up to 4 locations (aortic valve, pulmonic valve, left pulmonary artery, right pulmonary artery). Pertinent imaging parameters for clinical and rt-PC MRI are summarized in Table 1. Additional imaging parameters for rt-PC include radial k-space sampling with golden angle ratio = 111.246°.10

Image Reconstruction: As shown in Figure 1, we implemented a Cartesian GRASP framework by first shifting the polar data onto a Cartesian image space using NUFFT7, followed by estimating the Cartesian k-space pattern using gridding. Coil-combined zero-filled images, auto-calibrated sensitivity profiles11, and Cartesian k-space data were fed into the iterative CS framework during which artifacts were removed using both temporal finite difference and temporal principal component(PCA)12 as two orthogonal sparsifying transforms, where the regularization weight for PCA was 10 times smaller than that for temporal finite difference. The optimal normalized weights Cartesian GRASP, GRASP, and GROG-GRASP were determined empirically using visual analysis of training data set, and CS converged to a solution using 33 iterations of non-linear conjugate gradient with back-tracking line search. A single iteration using block matching (BM) filter9 was applied separately for both the magnitude and phase images to remove residual artifacts.

Data Analysis: To calculate forward flow, reverse flow, and peak velocity for a single representative heartbeat, region of interests were drawn manually on a workstation equipped with commercial software (CVi42, Circle Cardiovascular Imaging). Statistical analyses included ANOVA (Bonferroni with clinical as control), linear regression, and Bland-Altman.

Results

The mean reconstruction time was 229.5 min, 727.5 min, and 67.3 min for Cartesian GRASP, GRASP, and GROG-GRASP, respectively. Figure 2 shows the effects of increasing R on the Cartesian GRASP, GRASP, and GROG-GRASP reconstruction performance. Increasing acceleration from R = 4.8 to R = 38.4 resulted in considerable blurring for both GRASP and GROG-GRASP, whereas in Cartesian GRASP blurring was less pronounced. Figure 3, shows representative images as well as the corresponding flow and velocity curves of two different patients. In both instances, Cartesian GRASP produced better image quality and more accurate flow and velocity curves than GRASP and GROG-GRASP. Pooling data from 7 patients (N = 23 imaging planes), the mean forward and reverse flow values were not significantly different (p>0.48). For peak velocity, both GRASP and GROG-GRASP produced significantly lower (p<0.001) values, whereas Cartesian GRASP produced results that were not significantly different. For both forward and reserve flow values, comparing between clinical and rt-PC MRI, all three reconstruction methods produced similarly strong coefficient of determination (R2>0.93). For peak velocity, comparing between clinical and rt-PC MRI, Cartesian GRASP produced a higher coefficient of determination (R2=0.91) than GRASP (R2=0.27) and GROG-GRASP (R2=0.44). As summarized in Figure 4, Cartesian GRASP produced results that agree better with clinical than GRASP and GROG-GRASP.

Conclusion

The proposed rt-PC acquisition with 38.4-fold accelerated radial k-space sampling and Cartesian GRASP reconstruction method produced more accurate forward flow and peak velocity measurements in pediatric patients than GRASP and GROG-GRASP reconstruction frameworks.

Acknowledgements

This work was supported in part by the following grants:NIH R01HL116895, R01HL138578, R21EB024315, R21AG055954

References

1. Korperich H, Gieseke J, Barth P, Hoogeveen R, Esdorn H, Peterschroder A, Meyer H, Beerbaum P. Flow volume and shunt quantification in pediatric congenital heart disease by real-time magnetic resonance velocity mapping: a validation study. Circulation 2004;109(16):1987-1993.

2. Liu CY, Varadarajan P, Pohost GM, Nayak KS. Real-time color-flow MRI at 3 T using variable-density spiral phase contrast. Magn Reson Imaging 2008;26(5):661-666.

3. Lin HY, Bender JA, Ding Y, Chung YC, Hinton AM, Pennell ML, Whitehead KK, Raman SV, Simonetti OP. Shared velocity encoding: a method to improve the temporal resolution of phase-contrast velocity measurements. Magn Reson Med 2012;68(3):703-710.

4. Klein C, Schalla S, Schnackenburg B, Bornstedt A, Fleck E, Nagel E. Magnetic resonance flow measurements in real time: comparison with a standard gradient-echo technique. J Magn Reson Imaging 2001;14(3):306-310.

5. Lustig M, Donoho D, Pauly JM. Sparse MRI: The application of compressed sensing for rapid MR imaging. Magn Reson Med 2007;58(6):1182-1195.

6. Joseph A, Kowallick JT, Merboldt KD, Voit D, Schaetz S, Zhang S, Sohns JM, Lotz J, Frahm J. Real‐time flow MRI of the aorta at a resolution of 40 msec. Journal of Magnetic Resonance Imaging 2014;40(1):206-213.

7. Fessler JA. On NUFFT-based gridding for non-Cartesian MRI. J Magn Reson 2007;188(2):191-195.

8. Benkert T, Tian Y, Huang C, DiBella EVR, Chandarana H, Feng L. Optimization and validation of accelerated golden-angle radial sparse MRI reconstruction with self-calibrating GRAPPA operator gridding. Magn Reson Med 2018;80(1):286-293.

9. Haji-Valizadeh H, Collins JD, Aouad PJ, Serhal AM, Lindley MD, Pang J, Naresh NK, Carr JC, Kim D. Accelerated, free-breathing, noncontrast, electrocardiograph-triggered, thoracic MR angiography with stack-of-stars k-space sampling and GRASP reconstruction. Magn Reson Med 2018.

10. Winkelmann S, Schaeffter T, Koehler T, Eggers H, Doessel O. An optimal radial profile order based on the Golden Ratio for time-resolved MRI. IEEE transactions on medical imaging 2007;26(1):68-76.

11. Walsh DO, Gmitro AF, Marcellin MW. Adaptive reconstruction of phased array MR imagery. Magn Reson Med 2000;43(5):682-690.

12. Kim D, Dyvorne HA, Otazo R, Feng L, Sodickson DK, Lee VS. Accelerated phase-contrast cine MRI using k-t SPARSE-SENSE. Magn Reson Med 2012;67(4):1054-1064.

Figures

TABLE 1: Relevant imaging parameters. FB: free breathing; BH: breath-hold; VENC: velocity encoding strength.

Figure 1: A schematic of Cartesian GRASP reconstruction. In pre-processing, the original time series and time averaged polar k-space data were converted to images in Cartesian space using NUFFT. The polar k-space sampling mask was converted to Cartesian k-space sampling operator(F) using standard gridding. Auto-calibrated coil sensitivities(S) were derived from time averaged images as shown. The Cartesian time resolved images were transformed to k-space using FFT and then multiplied by k-space sampling operator to produce variable y, and the zero-filled, multi-coil, time-resolved images(x) were reconstructed using SENSE. During CS, temporal finite difference and temporal PCA were used as orthogonal sparsity transforms.

Figure 2: A) rt-PC images reconstructed at two different acceleration factors (R= 4.8 and R=38.4) from the same raw data (i.e. given Golden angles) using Cartesian GRASP, GRASP, and GROG-GRASP. At R=38.4, Cartesian GRASP produced considerably sharper images than GRASP and GROG-GRASP.

Figure 3: A) Image quality, and flow and peak velocity curves for clinical-PC MRI and rt-PC MRI reconstructed using Cartesian GRASP, GRASP, and GROG-GRASP at the right pulmonary artery level in an 11 year old patient. B) The corresponding data at the pulmonary valve level in a 5 year old patient.

Figure 4: Bland-Altman results for Cartesian GRASP, GRASP, and GROG-GRASP reconstructions: forward flow (left column), reverse flow (middle column), and peak velocity (right column). Both forward flow and peak velocity produced by Cartesian GRASP were in better agreement with those by clinical PC MRI than those produced by GRASP and GROG-GRASP. For reverse flow, given the small values, all three reconstructions showed similar agreement with clinical PC MRI. It should be noted that a mean difference of -1.5 mL in reverse flow is clinically insignificant.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)
4745