Efficient deep-learning-based reconstruction of Ferumoxytol-enhanced whole-heart 5D cardiac MRI

Motivation: Free-running cardiac magnetic resonance (CMR) imaging offers several advantages over conventional breath-held imaging, however the required compressed sensing (CS) reconstruction of this high-dimensional data is time-consuming, limiting its widespread clinical adoption. Goal(s): To develop a deep-learning-based reconstruction to rapidly obtain free-running CMR images of high quality. Approach: A modified residual neural network (FreeNet) is trained in a supervised manner on CS images to rapidly reconstruct free-running CMR data. Results: Our deep learning approach provides comparable 5D image quality to CS in less than one percent of the time required by CS. Impact: FreeNet demonstrates potential for rapid inline reconstruction of motion-resolved free-running CMR images for the first time. Our work is a preliminary step towards addressing current roadblocks, bringing "single-click" free-breathing CMR to wider patient populations.