INTRODUCTION

Cardiovascular magnetic resonance T₂ mapping enables quantification and visualization of changes in myocardial tissue(1). In addition, three-dimensional (3D) approaches,(2) remove limitations on volumetric coverage, maximum undersampling rate,(3) and achievable spatial resolution that can exist with two-dimensional approaches,(4). Nevertheless, the long scan time required for collection of different T₂-weighted 3D volumes can hinder the translation into clinical workflows. To reduce the scan time of 3D acquisitions while overcoming limitations of 2D approaches, we propose a fast single Breath-Holding 3D T2 mapping sequence (BH3DT2) with low-rank plus sparsity reconstruction.

METHODS

Imaging was performed on 3 T MR scanners (Ingenia CX, Philips Healthcare, Best, Netherlands). The human study was approved by the local institutional review board. Written informed consent was obtained from all subjects.
Sequence: The proposed BH3DT2 is based on our previously proposed free-breathing 3D T2 mapping sequence (FB3DT2), which acquires three interleaved ECG-triggered spoiled gradient echo volumes using T₂ preparation (T₂-prep) with different echo time (TE_T2Prep) under free breathing. A saturation pulse followed constant delay time is performed before acquisition each heartbeat to reset magnetization. Each volume was completed in 4 shots with a variable density spiral‐like Cartesian trajectory (5). Acquisition was completed within a single breath-holding of 12 heartbeats.
L+S Reconstruction: The T₂-weighted volumes can be represented by a Casorati matrix $$$L$$$ (6), with size $$$N×P$$$, where $$$N$$$ is the total number of voxels in 3D volume and $$$P$$$ is the number of distinct T₂-weighted volumes. The images share the same spatial structure, which can be represented as a low-rank matrix $$$L$$$, the residual matrix ($$$S$$$) between $$$X$$$ and $$$L$$$ is the result from the signal change of T₂ weighting contrast, which is assumed to be sparse. The imaging reconstruction was performed by solving the minimization problem: $$min⁡‖L‖_*+ λ‖TS‖_1 \ \ \ \ \ \ s.t.\ X=L+S,E(X)=d$$ where $$$‖L‖_*$$$ is the nuclear norm of the low-rank matrix $$$L$$$ , $$$T$$$ is the Fourier transform applied to $$$S$$$ to increase sparsity, and $$$‖TS‖_1$$$ is the l₁-norm. $$$λ$$$ is a parameter that balances the l1-norm and the nuclear norm. $$$E$$$ is the multicoil encoding operator, and $$$d$$$ is the undersampled multi-contrast multi-coil K-space data. The minimization problem was solved by iterative soft thresholding (7).
Retrospective Validation: Using data from one normal human subject and one swine with acute myocardial infarction, we investigate the optimal acceleration rate for BH3DT2 by retrospectively undersampling fully-sampled reference (Ref) data with three acceleration factors (R= 2,4,6). Reference data were acquired by FB3DT2 using the same signal preparation as BH3DT2 albeit with respiratory navigation. Imaging parameters of FB3DT2: FOV=300×300×80 mm³, voxel size=2×2×10 mm³, reconstructed resolution: 2×2×5 mm³, TE_T2Prep= 0, 25, 45 ms. Coil sensitivity map was estimated by sum of square. The normalized root-mean-square errors (nRMSE) were calculated to optimize the L+S reconstruction parameters, and to compare the performance.
Prospective Validation: Four normal human subjects (3 males, 28±7 years) were imaged with BH3DT2 and FB3DT2 in short axis orientation. Imaging parameters were: FOV=300×300×80mm³, voxel size=2×2×10mm³, TR/TE/ flip angle = 3.7ms/1.19ms/18°, 50 readouts per shot, acceleration factor R=4.48, TET2Prep=0, 25, 45 ms. Coil sensitivity map was estimated by ESPIRiT (8). FB3DT2 was performed without acceleration. The left ventricle was manually segmented on T2 maps and divided into 16 segments. The mean, standard deviation (SD), and coefficient of variation (CV) of T2 within each segment were calculated. A paired two-tailed Student's t-test was used for the statistical analysis.

RESULTS

Comparison results of the performance of L+S at different undersampling rates (R=2,4,6) are shown in Figures 1-3. With retrospective undersampling in both the normal human subject (Figure 1) and the swine with acute myocardial infarction (Figure 2), nRMSE increased with R. Image artefacts were subtle at R=4 though significantly increased in both T₂-weighted images and T₂ maps at R=6. Figure 3 indicates that there were little bias in mean T₂, and an increase in CV after undersampling. All subjects were scanned successfully with BH3DT2 (~12 seconds) and FB3DT2 (116±15 seconds) during prospective validation. Figure 4 shows T₂ measured with BH3DT2 was homogeneous over the whole heart, as demonstrated by the analysis in Figures 4B, 4C and 4D. Figure 5 shows the mean T₂ obtained from all four subjects over the whole left ventricle. The mean T₂ from BH3DT2 (47.2±1.2 ms) was only 1.5% higher than that from FB3DT2 (46.5±1.5 ms, p=0.03) with comparable variability (CV: 8.36±1.9% vs. 7.89±2.0%, p=0.02).

DISCUSSION and CONCLUSION

This study proposed and validated a 3D cardiac T₂ mapping sequence. In-vivo validations indicated that T₂ from breath-hold acquisitions (BH3DT2) had comparable accuracy to that from fully sampled data acquired during free-breathing (FB3DT2), with little impact on precision at high acceleration rates. In conclusion, the proposed sequence enables a fast estimate of T₂ across the whole heart within a single breath-hold. Further validation with additional subjects, and optimization of parameters for both acquisition and reconstruction are warranted.

Acknowledgements

This work was funded by National Key R&D Program of China (2016YFC0104700)