Harmonizer: A Deep Learning Framework for Transforming Low-resolution Clinical dMRI to Research-grade Quality

Motivation: Clinical diffusion MRI data often lack the spatial and angular resolution of research-grade data, limiting structural insights, especially in multi-site studies. Goal(s): To develop Harmonizer, a deep learning framework that enhances clinical diffusion MRI quality to research standards. Approach: Harmonizer combines Neural Orientation Distribution Field (NODF) for upsampling with CycleGAN for domain translation, aligning clinical data with research standards. Validation included both simulated and real-world clinical scans across multiple sites.. Results: Harmonizer improved clinical diffusion MRI quality, preserving sex-based FA differences in simulations and achieving a weighted-Dice score of 0.89 in vivo, demonstrating robust harmonization across datasets. Impact: his work enhances clinical dMRI data for multi-site neuroimaging studies, enabling high-quality, research-compatible analyses. The Harmonizer algorithm opens opportunities to study various disorders and pathologies using clinical data, supporting detailed white matter analyses and cross-site comparisons.