Synopsis

Spiral trajectories provide efficient data acquisition and favorable motion properties for cardiac MRI. We developed multiband (MB) methods to accelerate spiral cardiac cine imaging including a non-iterative spiral slice-GRAPPA (SSG) reconstruction and a temporal SSG (TSSG). Using 25-35% of k-space for single-band calibration data, experiments in phantoms and five volunteers show 18.7% lower mean artifact power than CG-SENSE when imaging three slices simultaneously. TSSG incorporating CAIPIRINHA with temporal alternation and a temporal filter in reconstruction further reduced rRMSE by 11.2% compared to SSG.

Purpose

Theory

The SSG method is illustrated in Figure 1, and the SSG reconstruction model can be expressed as follows:

$$x_s=SSG_sC\left( P_{s}^{*}\cdot X \right),$$

where the matrix $SSG_s$ is the spiral slice-GRAPPA kernel of the $s^{th}$ slice, $C$ is the gridding function, $P$ represents CAIPIRINHA phase modulation, $X$ is the multiband k-space data, and $x_s$ is the separated k-space data of slice $s$. As shown in Fig. 1A, the $SSG_s$ kernel is fitted using the single-band (SB) spiral center of k-space as calibration data. For this calculation, CAIPIRINHA phase modulation is applied to all slices, then phase demodulation corresponding to the $s^{th}$ slice is applied to all slices. Next, gridding is performed on all slices, and the split-slice GRAPPA method² is applied to fit the slice-GRAPPA kernel of the $s^{th}$ slice (Fig.1A). For image recovery, as shown in Figure 1B, the MB data are phase demodulated using the conjugate of the $s^{th}$ slice phase modulation matrix, $P_s^*$, and the gridding function $C$ is convolved with the MB data⁹. Next, the processed MB data are convolved with the $s^{th}$ slice-GRAPPA kernel, and the separated gridded k-space $(x_s)$ is obtained. Finally, the inverse Fast Fourier transform (IFFT) is performed to compute the image of the $s^{th}$ slice. This process is repeated for all slices. TSSG is based on alternation of CAIPIRINHA, and a temporal filter^{7, 8} is applied after SSG reconstruction.

Methods

Results

Figure 2 shows that rRMSE is minimized when 25-35% of the SB k-space are used for kernel calibration. Example images reconstructed using 15% (a-c) and 35% (d-f) of k-space for calibration are shown, as are corresponding artifacts relative to fully-sampled SB reference images (g-l). Panels (p) and (q) show the dependence of rRMSE on the spatial and temporal resolution of the calibration data. Based on these results, subsequent MB acquisitions used 35% of k-space and one cardiac phase for the SB calibration data.

Figure 3 shows phantom results comparing SSG and CG-SENSE for MB=3, where both methods used 35% of SB k-space for calibration. Less slice leakage artifact was achieved using SSG. Results from a volunteer are shown in Figure 4. Specifically, for a reference, fully-sampled SB images at basal, mid-ventricular and apical locations are shown in Figure 4(g-i), and CG-SENSE-recovered MB images (a-c) and SSG-recovered MB images (d-f) at the same locations are also shown. Red arrows indicate slice-leakage artifacts in CG-SENSE, and these are reduced using SSG. The artifact power⁴ of SSG was 18.7% lower than CG-SENSE (0.148±0.036 vs 0.182±0.037 for SSG vs. CG-SENSE, p<0.05, N=5). SSG required 30% of the computation time of CG-SENSE. Figure 5 compares results using SSG (Fig.5 (a-c)) and TSSG (Fig.5 (d-f)). The mean rRMSE of TSSG was 11.2% lower than SSG. The computation time for TSSG is similar to SSG.

Discussion

Acknowledgements

References