Accurate Peak and Mean Velocity with Deep Learning-Reconstructed highly Undersampled 4D Flow MRI using FlowVN

Motivation: 4D Flow MRI is limited by long scan times. Accelerated imaging by compressed sensing leads to long reconstruction times. Goal(s): The goal of this work was to evaluate a deep learning-based method (FlowVN) for reconstruction of pseudo-randomly heavily undersampled Cartesian 4D Flow. Approach: In this study, we explored FlowVN for the reconstruction of different acceleration factors and did velocity analysis of 4D Flow MRI. Results: The results show that FlowVN rapidly reconstructs undersampled 4D Flow images with good accuracy for average and peak velocity in the ascending aorta even at high acceleration factors. Impact: High-quality rapid reconstruction of highly undersampled 4D Flow MRI with deep learning has the potential to substantially facilitate the use of 4D Flow MRI in the clinical routine.