Motivation: Diffusion weighted MRI (DWI) captures information about the diffusion of water molecules by the application of strong bipolar gradients. It suffers from distortions in single-shot echo-planar imaging (ssEPI) due to B0 inhomogeneity and rapid signal decay. Multi-shot EPI (msEPI) enhances resolution but is affected by phase inconsistencies from motion artifacts. Goal(s): To tackle phase variation issues across shots in msEPI, this study introduces EPISTA+. Approach: The proposed approach unfolds the projection iteration algorithm1 into a deep network, integrating physical reconstruction models and image edge2 constraints for layer-wise recursive optimization. Results: Experimental results demonstrate EPISTA+ outperforms other deep learning methods in reconstructing high-quality images. Impact: This study introduced EPISTA+, a deep learning technique that integrates physical reconstruction model with image edge constraints to enhance the image quality of multi-shot DWI. This method improves image quality of high-resolution DWI and enhances diagnostic reliability in clinical settings.
Liu et al. (Tue,) studied this question.