Exploring SNR-Enhanced Reconstruction Techniques for Mesoscale Resolution BUDA-cEPI Diffusion MRI

Motivation: Diffusion MRI (dMRI) at mesoscale-resolution is hindered low SNR. To reconstruct high SNR images from high-noise levels and improve reconstruction schemes for clinical and research practice, we explore different advanced techniques to pinpoint optimal strategies. Goal(s): To investigate enhancing SNR reconstructions for mesoscale-resolution BUDA circular-EPI (BUDA-cEPI) dMRI. Approach: We employed BUDA-S-LORAKS, joint nearest-neighbors (JNN) diffusion directions with BUDA-S-LORAKS, and zero-shot self-supervised (ZS-SS) unrolled deep-learning network to reconstruct multiple directions from BUDA-cEPI at 500μm-resolution. Results: BUDA-cEPI reconstructions indicate that JNN BUDA-S-LORAKS and ZS-SS unrolled network with 1-NEX data achieve SNR levels comparable to standard method using 3-NEX data. Impact: Our exploration may provide the advancement of superb and appropriate dMRI reconstruction for high-fidelity dMRI at mesoscale resolution on clinical scanners by addressing SNR challenges and preserving fine anatomical details critical for accurate diagnosis and analysis.