4563

Cardiac Simultaneous Multi-Slice Multimapping based on Locally Low-Rank and Sparsity Constraints: Method Development and Validation

Yixin Emu¹, Zhuo Chen¹, Juan Gao¹, and Chenxi Hu¹
¹Shanghai Jiao Tong University, Shanghai, China

Synopsis

Keywords: Quantitative Imaging, Quantitative Imaging

Motivation: Simultaneous Multi-Slice (SMS) Multimapping can considerably improve the scan efficiency by acquiring myocardial T1 and T2 maps of 3 short-axis slices in 1 breath-hold.

Goal(s): To propose an accurate, precise, and reproducible SMS-Multimapping method.

Approach: A locally low-rank and sparsity-based reconstruction algorithm was developed to reduce noise and aliasing artifacts. Validation was performed in phantoms and 10 healthy subjects, where the method was compared with standard MOLLI/bSSFP T2 mapping and Multimapping.

Results: Despite the 6-fold reduction of scan time, the proposed method shows good accuracy, reasonable precision, and acceptable reproducibility in its regional myocardial T1 and T2 measurement.

Impact: The proposed method transforms a scan of 6 breath-holds down to a single breath-hold scan. Once employed, the method can greatly improve the patient comfort and efficiency of myocardial parametric mapping.

Introduction

Cardiac T1 and T2 mappings are clinically used to quantitatively evaluate myocardial fibrosis and edema^1–3. However, the application of T1 and T2 mappings in 3 short-axis slices requires at least 6 breath-holds. Multimapping⁴ has been recently developed to acquire T1 and T2 maps in one breath-hold. A combination of Simultaneous Multi-Slice (SMS)⁵ and Multimapping can further reduce the scan time to one breath-hold; however, the combined in-plane and through-plane accelerations could cause a severely ill-conditioned reconstruction problem^6–9. Here we propose an SMS-Multimapping sequence and a Locally Low-Rank (LLR)^10–12 and Sparsity^12,13 based reconstruction algorithm LLRS to both improve the scan efficiency and maintain the mapping quality. We compared the method with MOLLI, bSSFP T2 mapping, and Multimapping in 10 healthy subjects in terms of regional accuracy, standard deviation, and reproducibility.

Methods

Fig. 1A shows the design of the SMS-Multimapping sequence. Similar to Multimapping, the sequence uses 10 heartbeats to generate 10 images of variable T1 and T2 weightings. However, unlike Multimapping, the SMS-Multimapping sequence uses FLASH readouts with multiband RF excitations to excite 2-3 slices simultaneously. Fig. 1B shows a schematic of the k-space sampling trajectory, which is based on the previously published SUPER-CAIPIRINHA¹⁴ sampling pattern. For multiband factors of 1, 2, and 3, the total acceleration rate is roughly 4, 8, and 12, respectively. The cost function for the underlying reconstruction problem is formulated as a sum of the data-fidelity term, LLR term, and sparsity term:

$$\min _{\boldsymbol{x}} \frac{1}{2}\|\boldsymbol{y}-\boldsymbol{D F} \boldsymbol{E} \boldsymbol{x}\|_F^2+\alpha \sum_{i=1}^{N_v}\left\|\boldsymbol{B}_{i} \boldsymbol{x}\right\|_*+\beta\|\boldsymbol{W} \boldsymbol{x}\|_1$$

where $$$\boldsymbol{y}$$$ is the acquired k-space data, $$$\boldsymbol{D}$$$ the undersampling operator, $$$\boldsymbol{F}$$$ the Fourier transform, $$$\boldsymbol{E}$$$ the coil sensitivity encoding, $$$\boldsymbol{x}$$$ the multislice multi-contrast images, $$$\left \| ⋅ \right \|_{F}$$$the matrix Frobeneous norm, $$$\left \| ⋅ \right \|_{*}$$$ the nucleus norm, $$$\alpha$$$ and $$$\beta$$$ the regularization weights, $$$W$$$ is the wavelet transform, and $$$\boldsymbol{{B}} _{i}$$$ the block-extraction operator, which takes a block of voxels centered at the ith voxel to form the Casorati matrix. The above inverse problem is solved by the ADMM¹⁵ algorithm. After the reconstruction of $$$x$$$, mapping is performed by dictionary matching. The dictionary is generated by Bloch Equation simulation of the SMS-Multimapping sequence for a T1 range of [200ms:1ms:2500ms] and a T2 range of [1ms:1ms:150ms].

SMS-Multimapping was validated via phantom and in vivo imaging. Scans were performed in a 3T scanner (uMR790, United Imaging Healthcare, Shanghai, China) with a 24-channel spine coil and 12-channel torso coil. Ten healthy subjects (4 females, age 23±3 years) were imaged after the subjects provided written informed consent. Among the 10 subjects, 3 had a repeated scan a week later, enabling a preliminary evaluation of the scan-rescan reproducibility. Acquisition parameters for SMS-Multimapping were FOV/matrix size/flip angle/TR/TE/bandwidth/slice thickness/acquisition window=360×270mm²/192×138/5°/4.10ms/1.73ms/400Hz/pixel/8mm/201ms.

Results

Fig. 2A shows the phantom reconstructions. Visual inspection showed a consistent quality between multiband factor=1 vs multiband factor=3. Fig. 2B shows that the T1 and T2 values measured by SMS-Multimapping (MB=3) agreed well with those measured by spin echo imaging, although errors slightly increased for larger T1, T2, and heart rates.

Fig. 3 shows the in vivo reconstruction of one subject. The use of LLRS reconstruction substantially reduced the noise and aliasing artifacts compared with SPSG+GRAPPA⁵, which was commonly used in other SMS applications⁵.

Fig. 4 shows the comparisons between MOLLI/T2 mapping, Multimapping⁴, and SMS-Multimapping (MB=3) in terms of visual qualities (A), regional T1/T2 means (B), and regional T1/T2 standard deviations (C) over 10 subjects. SMS-Multimapping showed an overall similar quality relative to Multimapping. Both Mutlimapping and SMS-Mutlimapping showed higher regional T1s relative to MOLLI and similar regional T2s relative to T2 mapping. SMS-Multimapping had an overall lower T1 precision relative to MOLLI and Multimapping, and a lower T2 precision relative to T2 mapping.

Fig. 5 shows preliminary evaluations of the regional T1/T2 scan-rescan reproducibility in 3 subjects. The regional T2 of SMS-Multimapping appeared very reproducible, while the regional T1 of SMS-Multimapping appeared less reproducible than the other methods.

Discussion and Conclusions

SMS-Multimapping is able to fulfill simultaneous 3-slice T1 and T2 mapping in 10 heartbeats, which is a reasonable time for a single breath-hold. The proposed LLRS reconstruction well mitigated the noise and artifacts caused by the acceleration. The regional T1 and T2 measured by SMS-Multimapping showed good accuracy, albeit the precision and reproducibility appeared slightly lower than MOLLI/T2 mapping, potentially due to the inherent lower SNR of the FLASH readout and the large acceleration rate. The 6-fold reduction of scan time could greatly improve patient comfort and make parametric mapping more feasible for severely ill patients in clinical practice.

Acknowledgements

No Acknowledgements found.

References

1. Taylor AJ, Salerno M, Dharmakumar R, Jerosch-Herold M. T1 Mapping: Basic Techniques and Clinical Applications. JACC Cardiovasc Imaging. 2016;9(1):67-81. doi:10.1016/j.jcmg.2015.11.005

2. Kellman P, Aletras AH, Mancini C, McVeigh ER, Arai AE. T2-Prepared SSFP Improves Diagnostic Confidence in Edema Imaging in Acute Myocardial Infarction Compared to Turbo Spin Echo. Magn Reson Med. 2007;57(5):891-897. doi:10.1002/mrm.21215

3. Kellman P, Wilson JR, Xue H, Ugander M, Arai AE. Extracellular volume fraction mapping in the myocardium, part 1: evaluation of an automated method. J Cardiovasc Magn Reson. 2012;14(1):63. doi:10.1186/1532-429X-14-63

4. Henningsson M. Cartesian dictionary-based native T1 and T2 mapping of the myocardium. Magn Reson Med. 2022;87(5):2347-2362. doi:10.1002/mrm.29143

5. Barth M, Breuer F, Koopmans PJ, Norris DG, Poser BA. Simultaneous multislice (SMS) imaging techniques. Magn Reson Med. 2016;75(1):63-81. doi:10.1002/mrm.25897

6. Weingärtner S, Moeller S, Schmitter S, et al. Simultaneous multislice imaging for native myocardial T1 mapping: Improved spatial coverage in a single breath-hold. Magn Reson Med. 2017;78(2):462-471. doi:10.1002/mrm.26770

7. Ye H, Cauley SF, Gagoski B, et al. Simultaneous multislice magnetic resonance fingerprinting (SMS-MRF) with direct-spiral slice-GRAPPA (ds-SG) reconstruction. Magn Reson Med. 2017;77(5):1966-1974. doi:10.1002/mrm.26271

8. Hilbert T, Schulz J, Marques JP, et al. Fast model-based T2 mapping using SAR-reduced simultaneous multislice excitation. Magn Reson Med. 2019;82(6):2090-2103. doi:10.1002/mrm.27890

9. Gaspar AS, Silva NA, Price AN, Ferreira AM, Nunes RG. Open-source myocardial T1 mapping with simultaneous multi-slice acceleration: Combining an auto-calibrated blipped-bSSFP readout with VERSE-MB pulses. Magn Reson Med. April 2023. doi:10.1002/mrm.29661

10. Zhang T, Pauly JM, Levesque IR. Accelerating parameter mapping with a locally low rank constraint: Locally Low Rank Parameter Mapping. Magn Reson Med. 2015;73(2):655-661. doi:10.1002/mrm.25161

11. Mao X, Lee HL, Hu Z, et al. Simultaneous Multi-Slice Cardiac MR Multitasking for Motion-Resolved, Non-ECG, Free-Breathing T1–T2 Mapping. Front Cardiovasc Med. 2022;9:833257. doi:10.3389/fcvm.2022.833257

12. Hamilton JI, Jiang Y, Ma D, et al. Simultaneous multislice cardiac magnetic resonance fingerprinting using low rank reconstruction. NMR Biomed. 2019;32(2):e4041. doi:10.1002/nbm.4041

13. Wang X, Rosenzweig S, Scholand N, Holme HCM, Uecker M. Model-based reconstruction for simultaneous multi-slice mapping using single-shot inversion-recovery radial FLASH. Magnetic Resonance in Medicine. 2021;85(3):1258-1271. doi:10.1002/mrm.28497

14. Yang F, Zhang J, Li G, Zhu J, Tang X, Hu C. Regularized SUPER-CAIPIRINHA: Accelerating 3D variable flip-angle T1 mapping with accurate and efficient reconstruction. Magn Reson Med. 2023;90(4):1380-1395. doi:10.1002/mrm.29714

15. Boyd S. ADMM. FNT in Machine Learning. 2010;3(1):1-122. doi:10.1561/2200000016

Figures

Figure 1.(A) A schematic of the SMS-Multimapping sequence, which uses multiband RF pulses and FLASH readouts to acquire the SMS data. (B) The sequence samples k-space based on the recently proposed SUPER-CAIPIRINHA pattern. This pattern includes accelerations along both the in-plane and through-plane directions, among which the in-plane acceleration was shifted at every contrast. In this work, the total undersampling rates of k-space for 1, 2, and 3 slices are 4, 8, and 12, respectively.

Figure 2. (A) The reconstructed phantom SMS-Multimaps for Multiband (MB) factors of 1 and 3 at an 80bpm heart rate. The T1 and T2 maps at MB=3 were in a good agreement with those acquired slice-by-slice (MB=1). (B) The comparison of T1 and T2 values measured by 3-slice SMS-Multimapping and those measured by spin echo imaging for different vials and heart rates. SMS-Multimapping showed a good agreement with reference values for both T1 and T2, although errors slightly increased for larger T1, T2, and heart rates.

Figure 3. Reconstructions of in vivo SMS-Multimaps by SPSG+GRAPPA, a common approach for SMS reconstructions, and the proposed LLRS method. For MB=2 and 3, LLRS substantially reduced noise and aliasing artifacts compared with SPSG+GRAPPA. The quality of LLRS at MB=3 was overall agreeable with that of LLRS at MB=1.

Figure 4. (A) T1 and T2 maps obtained by MOLLI/T2 mapping (6 breath-holds), Multimapping (3 breath-holds), and SMS-Multimapping (1 breath-hold). Bullseye plots of the regional means (B) and Standard Deviations (SD) (C) over 10 subjects. Both Multimapping and SMS-Multimapping caused statistically longer T1s in most segments than MOLLI. Regional T2s were similar between the three methods. SMS-Multimapping had a higher T1 SD than others, and a higher T2 SD than T2 mapping. The symbol * represents P<0.05.

Figure 5. The Bland-Altman analysis of regional T1s and T2s measured by repeated MOLLI, T2 mapping, Multimapping, and SMS-Multimapping scans (one-week apart) from 3 healthy subjects. Only the 6 myocardial segments in the mid-ventricular slice were considered. The regional T2 measured by SMS-Multimapping was very reproducible, while the regional T1 measured by SMS-Multimapping appeared less reproducible than the other two methods.

Proc. Intl. Soc. Mag. Reson. Med. 32 (2024)

4563

DOI: https://doi.org/10.58530/2024/4563