Changyu Sun1, Kenneth C. Bilchick2, Daniel Weller3, 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
Synopsis
k-t
undersampled multiband first-pass perfusion imaging provides in-plane and
through-plane acceleration and potentially enables high spatial resolution and
whole-heart coverage. Our slice-L+S work splits MB data consistency and enforcing L+S into two subproblems wherein MB data,
in- and through-plane coil information and L+S
properties of slices are systematically utilized. Here we compare slice-L+S to
single-band L+S and slice-GRAPPA followed sequentially by L+S (seq-SG-L+S)
using 3 retrospectively undersampled datasets. High resolution nine-slice data
with multiband factor-3 and rate-3 undersampling were prospectively acquired in
4 patients. Slice-L+S showed better image
quality demonstrated by quantitative results and scores from two cardiologists.
Purpose
First-pass MRI is routinely
used to image myocardial perfusion1. However, achieving high
spatial-temporal resolution, high slice coverage and high signal-to-noise-ratio
is challenging2,3. k-t undersampled multiband (MB) methods may meet
these challenges, and for this purpose we recently developed a slice-low rank
plus sparse (slice-L+S) reconstruction model2 that integrates MB
data consistency4 with low-rank plus sparse constraints. The purposes of the present study were to (a)
compare MB slice-L+S with k-t undersampled single-band (SB) first-pass imaging using
an L+S reconstruction and equal total acceleration factor, and (b) compare the
integrated slice-L+S reconstruction method with the sequential operations of split
slice-GRAPPA followed by single-band L+S (seq-SG-L+S) for the reconstruction of
k-t undersampled MB first-pass data. We focused on Cartesian imaging because it
is the most widely used trajectory and we targeted rate-6-9 acceleration in
order to cover 6-9 slices with 1.5mm×1.5mm in-plane resolution. Methods
A saturation-recovery gradient-echo sequence
was modified to use MB excitation with CAIPIRINHA5 phase modulation
and Poisson-disk k-t undersampling. SB
calibration data were acquired in the first heartbeat, and were used to
calibrate ESPIRiT6 maps and split slice-GRAPPA kernels7 for
each slice. Fully-sampled SB first-pass perfusion images were similarly
acquired and used to retrospectively synthesize various k-t undersampled MB datasets
(Figure 1). Imaging was performed on a
1.5T system (Aera, Siemens) using 20-34 receiver channels. Fully-sampled SB datasets
from 3 patients were acquired using 3 slices and a spatial resolution of 3.3mm×3.3mm. To
compare MB slice-L+S with SB L+S using an equal total acceleration factor, we retrospectively
synthesized k-t undersampled MB and SB data from the fully-sampled SB data
(Figure 1). The MB slice-L+S reconstruction
used a MB factor of 3 (MB=3) and k-t underampling factors of 2 and 3 (R=2 and R=3),
for total accelerations of 6 and 9. The SB
L+S reconstruction used R=6 and 9 (MB=1). To compare the different MB reconstruction
methods (slice-L+S vs seq-SG-L+S), we again retrospectively synthesized data
and we additionally acquired 4 prospectively undersampled MB first-pass
datasets. These acquisitions used MB=3,
R=3, spatial resolution of 1.5mm×1.5mm,
nine total slices, and a temporal footprint of 169ms per slice. The results of the
MB (MB=3, R=2 and R=3) vs SB (R=6 and 9) comparisons were quantified using normalized
root mean squared error (nRMSE) and structural similarity (SSIM). The results of the slice-L+S and seq-SG-L+S reconstructions
of synthesized data were quantified using nRMSE and SSIM, and those of the prospectively
acquired undersampled MB data were scored by two
cardiologists (1-5, 5 is best). Results
For the comparison of MB
slice-L+S and SB L+S using an equal total acceleration factor, Figure 2 shows example reconstructions using both methods
for total accelerations of rate 6 and rate 9.
MB slice-L+S outperformed SB L+S in both cases. For rate 6, the mean nRMSE and SSIM of slice-L+S
were 23% lower and 2.2% higher than SB L+S. At rate-9, the mean nRMSE and SSIM
of slice-L+S were 27% lower and 4.8% higher than SB L+S. For the comparison of MB slice-L+S and MB seq-SG-L+S,
Figure 3 shows example reconstructions using both methods for total accelerations
of rate 6 (MB=3, R=2) and rate 9 (MB=3, R=3), demonstrating that slice-L+S
produced lower overall artifacts. Example images and movies demonstrating the better
image quality of slice-L+S vs seq-SG-L+S with MB=3 and R=3 are shown in Figures
4 and 5, demonstrating 9-slice coverage with 1.5mm×1.5mm spatial resolution. At rate-6, the mean nRMSE and SSIM of slice-L+S
were 25% lower and 4.6% higher than seq-SG-L+S. At rate-9, the mean nRMSE and
SSIM of slice-L+S were 32% lower and 8.6% higher than seq-SG-L+S. The cardiologist scoring results were 2.6 ± 0.58
vs 4.1 ± 0.73 (p<0.001) for seq-SG-L+S and slice-L+S, respectively, with
very close agreement noted between observer ratings for the images (r=0.92,
p=0.001).Discussion
These results suggest that slice-L+S for the
reconstruction of k-t undersampled MB first-pass images provides better image
quality than L+S reconstructions of SB first-pass images with the same total
accelerations of rate 6-9. Further, for
undersampled MB first-pass imaging, the integrated slice-L+S method outperforms
the sequential operations of split-slice-GRAPPA and L+S. Slice-L+S with MB=3 and R=3 at 1.5T provides
high spatial resolution (1.5mm×1.5mm)
and nine slices, as well as good image quality as judged by expert readers. In
the future, additional datasets will be acquired and analyzed to confirm the
present results. Acknowledgements
This work was
supported by R01HL147104.References
- Kramer CM, Barkhausen J,
Flamm SD, Kim RJ, Nagel E, Society for Cardiovascular Magnetic Resonance Board
of Trustees Task Force on Standardized P. Standardized cardiovascular magnetic
resonance (CMR) protocols 2013 update. J Cardiovasc Magn Reson 2013;15:91.
- Sun C, Robinson A, Schumann
C, Salerno M, Epstein FH. Multiband first-pass perfusion MRI using
slice-low-rank plus sparse. ISMRM 2020:1097.
- Otazo R, Candes E,
Sodickson DK. Low-rank plus sparse matrix decomposition for accelerated dynamic
MRI with separation of background and dynamic components. Magn Reson Med
2015;73(3):1125-1136.
- Sun C, Yang Y, Cai X, Salerno M, Meyer CH, Weller D, Epstein FH. Non-Cartesian slice-GRAPPA and slice-SPIRiT reconstruction methods for multiband spiral cardiac MRI. Magn Reson Med 2020;83(4):1235-1249.
- Breuer FA, Blaimer M,
Heidemann RM, Mueller MF, Griswold MA, Jakob PM. Controlled aliasing in
parallel imaging results in higher acceleration (CAIPIRINHA) for multi-slice
imaging. Magn Reson Med 2005;53(3):684-691.
- Uecker M, Lai P, Murphy MJ,
Virtue P, Elad M, Pauly JM, Vasanawala SS, Lustig M. ESPIRiT--an eigenvalue
approach to autocalibrating parallel MRI: where SENSE meets GRAPPA. Magn Reson
Med 2014;71(3):990-1001.
- Cauley SF, Polimeni JR,
Bhat H, Wald LL, Setsompop K. Interslice leakage artifact reduction technique
for simultaneous multislice acquisitions. Magn Reson Med 2014;72(1):93-102.