Accelerated 4D flow imaging with variable-density cartesian undersampling and parallel imaging reconstruction

Methods Variable-density Poisson Disk Distribution (VD-PDD) undersampling was applied for 4D flow imaging. By applying VD-PDD independently at each time frame (Figure 1a), we achieved random undersampling in both the ky-kz plane and temporal domains. In addition, we applied an improved initial solution for SPIRIT to significantly improve reconstruction accuracy and robustness. We explored the different undersampling and reconstruction algorithms on fully sampled 4D flow CMR data acquired on a 1.5T Siemens Avanto scanner with a 5-ch coil in 3 volunteers (venc=200 cm/s, FOV=320x240x55 mm, matrix=128x96x22, ~18 time frames of 35 ms temporal resolution, ~25 mins scan time). VD-PDD (center 12x12 fully sampled; R=6) was retrospectively applied to the full data. Composite data was generated by sharing data from other frames based on the temporal distance to the time frame of interest (Figure 1b). k-space was filled to the extent possible from selected adjacent time frames. The images reconstructed from the composite data through time (Figure 1c) were used as initial solutions for SPIRIT. This is denoted as method “M1”. We also generated composite data that shared limited data from other frames (Figure 1d) to mitigate undersampling of the data. Then SPIRIT with the proposed initial solutions was applied to this new composite data through time (Figure 1e), referred to as “M2”. Flow-waveforms in the ascending (AA) and descending aorta (DA) were measured in 5 locations in each subject (Figure 2d-e). Relative error was calculated with the fully sampled data as reference.