K-t PCA accelerated in-plane balanced steady-state free precession phase-contrast (PC-SSFP) for all-in-one diastolic function evaluation

FIGURE S1. Estimated acceleration and induced phase error in vivo and in phantom. Since gradient moments of higher order (e.g., M_x2) were not nulled for these PC approaches, and since the moments differed for the PC-GRE and PC-SSFP approaches in sign and slightly in magnitude, acceleration might induce differing errors in velocity mapping for these approaches, resulting in worse agreement if acceleration is high. Because the circular flow phantom had very high accelerations, a_x, when v_x was lowest, and because the 2nd moments differed between PC-GRE and PC-SSFP, the impact of acceleration can be observed in Figure 3B. However, the accelerations in the phantom were four-fold greater than we estimated in vivo, where the extra phase from M_x2 will be generally very small (<1%) compared to M_x1.

FIGURE S2. Comparison between sequential and interleaved PC-SSFP. Reference and velocity encoded readouts were executed sequentially to maintain their respective steady states. See the additional banding artifacts, using interleaved acquisition, in both magnitude and phase images (blue arrow).

FIGURE S3. One example showing the separate reference and velocity-encoded phase images. Blood velocities could potentially be obtained from the velocity-encoded data alone, which provided similar flow curves to the conventional subtractive approach.

FIGURE S4. Intro-subject scan-rescan reproducibility of PC-SSFP. (A) Measurements of E, A, e′ operated at a different Larmor frequency (optimal and suboptimal frequency offset). Coefficients of variance were all reasonable. (B) If the blood pulse sits in the dark bands, the measured velocity cannot be trusted. See how the phases were affected by the bands, and how they were recovered after shifting away the bands.

FIGURE S5. Comparison between global and compartment-based k-t PCA. Residual aliasing from compartment-based approaches was found by a greater chance compared to global k-t PCA methods. The artifacts affected both intensity and phase images.

FIGURE S6. Comparison between 8 k-t PCA methods. Percentage difference between prospectively undersampled k-t PCA PC SSFP with high temporal resolution, and standard PC-SSFP. Left matrix of each pair contains values without retained raw data while the right matrix substituted it to the k-t PCA synthetic k-space, each grid inside representing one method as demonstrated in the lower right corner, indicating percentage difference mean and its standard deviation. With retained raw k-space, E and A peaks were recovered, having smaller difference (p < 0.001), and E/A ratios were lower (p < 0.001), in all methods. However, e′ peaks showed larger variance among different methods (especially considering that one bad-tracked data could give a completely different e′). In our study, compartment-based methods were less reliable (larger variance) and should be avoided. Meanwhile, independent and complex difference approaches provided similar results, while the later one had better performance on e′ evaluation. Therefore, global k-t PCA with complex difference input (green box) would be suggested.