Advancing Multishot EPI Reconstruction for Diffusion MRI: From Locally to Non-Locally Low-Rank Methods

Motivation: High-SNR, high-resolution diffusion MRI is essential for detailed anatomical imaging. To this end, multishot EPI (ms-EPI) can be helpful but is challenged by low SNR and motion-induced phase variations. Goal(s): This study aims to develop a new non-locally low-rank reconstruction (NLLR) method that corrects motion-induced shot-to-shot phase variations and improves noise suppression in DW ms-EPI, improving high-resolution DWI. Approach: We extend the shot-locally low-rank (LLR) method by incorporating spatially non-local patch similarities into ms-EPI reconstruction that improves low-rank approximations and noise reduction. Results: NLLR achieved higher structural clarity and higher SNR compared to the LLR method in both simulations and in-vivo validations. Impact: This study presents a robust non-local-low-rank reconstruction technique for multi-shot EPI in diffusion MRI that achieves superior image quality and SNR. These advances provide clearer anatomical details, supporting improved diagnosis and encouraging the exploration of non-local approaches in high-resolution imaging.