Comparison of reconstruction methods of highly accelerated 3D dynamic cardiovascular MRI using the 4D XCAT phantom

Motivation: Reconstruction of highly accelerated cardiovascular MRI is critical, but it is difficult to obtain a standard ground truth for comparing different methods because of the respiratory and cardiac motion. Goal(s): Acquiring a fully-sampled 4D cardiovascular simulation data and comparing different reconstruction methods. Approach: Using the phantom generated by XCAT, we compared locally low-rank (LLR), low-rank plus sparse (L+S) and haar wavelet (HAAR) reconstruction methods at 8×~24× under-sampling rates. Results: HAAR achieved the best performance across all the comparison metrics and under-sampling rates. Impact: A numerical phantom has been provided, along with the implementation and comparison of three different reconstruction methods. The code is available at https://github.com/AssociatedPrimeIdeal/MRI_TorchRecon