Free-Running Self-Gated High-Resolution Volumetric Multi-Slab Cardiac Cine MRI at 3T

Background: Free-running self-gated 3D cine is attractive for volumetric, high-resolution, breath-hold-free assessment of LV function, yet implementation at 3T is constrained by high SAR for bSSFP and low myocardium-blood contrast in both bSSFP and GRE [1-3]. To address these issues, we develop a multi-thin-slab 3D GRE cine approach for 3T high-resolution whole-heart imaging with eddy-current-robust sampling and motion-corrected reconstruction.

Methods: A free-running 3D GRE with pseudo-radial Cartesian sampling (Fig. 1A) was implemented. Each interleaf had two out-in-out variable-density spokes separated by golden-angle to reduce eddy currents. Central k-space PCA provided motion binning signals (Fig. 1B). Respiratory motion was corrected as slab-wise translation via k-space phase correction. Cardiac motion-resolved images were reconstructed by 2D LLR SENSE. Cost function:

min_x 1/2‖W(Ex−My)‖_F^2+α∑_i‖B_i(x)‖_* (1)

where x is the reconstructed 4D image; y the k-space data; E the encoding matrix; W the view-sharing weights; M the motion matrix in k-space; α the low-rank parameter; B_i the voxel-block extractor; and ‖·‖_* the nuclear norm. The cost function was minimized by ADMM [4, 5].

Imaging: scans on a 3T scanner (uMR 790, United Imaging Healthcare, Shanghai, China) with torso and spine coils. 15 volunteers (3 female, 22.6 ± 2.0 y) underwent proposed multi-slab 3D cine and reference 2D breath-hold cine; 10 also had single-slab 3D. Image quality (SNR, CNR, myocardial sharpness, residual artefact) and LV volumes (EDV, ESV, EF) were compared. Proposed 3D parameters: FOV 285×285×120 mm³, resolution 1.98×1.98×1.5 mm³, 30% slice-direction oversampling, slab thickness 30 mm, flip 10°, TR/TE 4.33/1.44 ms, bandwidth 910 Hz. 2D bSSFP: in-plane 1.98×1.98 mm², slice 8 mm, flip 50°, TR/TE 2.97/1.40 ms, bandwidth 1000 Hz.

Results: Fig. 2 shows ED and ES images for the multi-slab 3D GRE, single-slab 3D GRE, and 2D bSSFP cine. The proposed method improved image quality over single-slab 3D cine (sharpness 2.82±0.42 vs 1.58±0.46; artefact 2.58±0.26 vs 1.13±0.21; both P=0.005).

Fig. 3 shows Bland-Altman and linear regression for EDV, ESV and EF measured by multi-slab 3D and 2D cine. Multi-slab 3D cine yielded comparable LV measurements with 2D cine (EDV 140.3±19.9 vs 139.3±20.0 mL, P=0.357, r=0.976; ESV 56.7±11.8 vs 57.7±10.2 mL, P=0.259, r=0.964; EF 59.8%±4.9% vs 58.7%±3.7%, P=0.073, r=0.907). Although contrast could decline at end-systole in apical slices, and may vary across slabs due to different flow rates, the retained image contrast sufficed for LV assessment with acceptable accuracy.

Conclusion: A free-running, self-gated multi-slab 3D GRE cine framework enables whole-heart 3T cine with 1.5-mm through-plane resolution. It improves image quality over single-slab cine, and agrees well with breath-hold 2D cine for LV volumes in the preliminary investigation. The technique can be useful when high spatial resolution, free-breathing and ECG-free acquisitions are desired.

Figure 1. Illustration of the sampling strategy and motion binning for the proposed method.
Figure 2. Reconstruction results of the proposed multi-slab 3D cine, single-slab 3D cine and breath-hold multi-slice 2D cine in a representative subject. Both short-axis and long-axis (2-chamber and 4-chamber) images in end-diastole and end-systole are shown. Notice the improved image contrast of the proposed method relative to the single-slab method, and the improved through-plane resolution relative to the 2D breath-hold multi-slice method. The retained image contrast (even with inversion) enables delineation of the endocardial and epicardial borders.
Figure 3. (A)-(C) Bland-Altman plots of the end-diastolic volume (EDV), end-systolic volume (ESV) and LV ejection fraction (EF) for the proposed multi-slab 3D cine and breath-hold 2D cine. Mean bias and 95% confidence intervals are displayed in the figure. Biases were small compared with the actual values of these metrics, suggesting accurate assessments of LV function by the proposed method despite the inferior image contrast and image quality. (D)-(F) Linear regression of EDV, ESV and LVEF between the proposed multi-slab 3D cine and gold-standard breath-hold 2D cine. Dashed red line is the identity line, and the solid line is the regression line.