Spatio-Angular Noise2Noise for Self-Supervised Denoising of Diffusion MRI Data

Motivation: Diffusion MRI (DMRI) suffers from heavy noise. The noise issue reduces the accuracy and reliability of the derived diffusion metrics. Goal(s): Existing Deep Learning (DL) methods for DMRI denoising usually rely on training with paired noisy-clean data, which are unavailable in a clinical setting. Therefore, we propose a self-supervised DL denoising method, called Spatio-Angular Noise2Noise, for DMRI denoising. Approach: We stem from the fact that a network trained with paired noisy data can capture the essential information of underlying clean data for noise reduction. Results: Extensive experiments on simulated and real datasets demonstrate the superiority of SAN2N over existing DMRI denoising methods. Impact: SAN2N can reduce the noise effectively and improve the quality of fiber ODFs and tractography.