Accelerating Free-Breathing Liver MRI at 7T using Recurrent Inference Machines

Motivation: Liver imaging could benefit from high SNR and spatial resolution offered at 7T but suffers from B1+ inhomogeneities and lengthy acquisitions during breathing. Goal(s): To improve the reconstruction accuracy for prospectively undersampled free-breathing liver data while reducing processing time. Approach: Free-breathing, rf-shimmed imaging reconstructed with CIRIM for rapid image reconstruction at high undersampling factors. Results: CIRIM reconstructed breath-hold liver data for acceleration factors up to 6 without significant loss of detail and was successfully applied to prospectively undersampled data acquired at isotropic resolutions of 1.5mm and 1.35mm. Impact: This work advances ultra-high field, free-breathing liver MRI with deep-learning reconstruction, offering improved motion robustness over CS. It paves the way for prospectively undersampled, submillimeter resolution free-breathing acquisitions in future studies while maintaining short scan durations and minimizing patient discomfort.