FlowMRI-Net: A generalizable self-supervised physics-driven 4D Flow MRI reconstruction network for aortic and cerebrovascular applications

Motivation: Improved reconstruction quality and speed is necessary to accelerate 4D flow MRI acquisition and promote clinical adaptation. Goal(s): To develop a deep learning-based framework (FlowMRI-Net) for fast reconstruction of accelerated 4D flow MRI that can be used for applications where reference data are not available. Approach: Training is performed in a self-supervised manner using healthy aortic and cerebrovascular acquisitions and results are compared to state-of-the-art compressed sensing and deep learning-based (FlowVN) methods. Results: FlowMRI-Net outperforms CS-LLR and FlowVN for aortic 4D flow MRI reconstruction and CS-LLR for cerebrovascular 4D flow MRI reconstruction. Impact: FlowMRI-Net facilitates higher undersampling factors than the current state-of-the-art for aortic and cerebrovascular 4D flow MRI within clinically feasible reconstruction times, improving clinical adaptation particularly for cerebrovascular applications which are otherwise too time-consuming.