Denoising diffusion MRI data: Principal components meet non-local block-matching

Motivation: Diffusion MRI suffers from very low SNR, especially for high spatial resolution and/or diffusion gradient strength (high b-values). Goal(s): We aim to mitigate the noise in dMRI, improve over existing denoising algorithms, and suppress noise while preserving tissue details. Approach: We proposed a novel denoising method for dMRI that combines global singular value decomposition and non-local block-matching denoising for the Principal Components. Results: Our method outperformed several state-of-the-art algorithms by improving DWI image quality, providing noise suppression without signal loss, and preserving tissue features. Furthermore, our results achieved higher accuracy in estimating diffusion metrics such as FA, NDI, and excess kurtosis. Impact: The innovative BM4PC reduces noise in diffusion-weighted images, enabling diagnostic capabilities and sensitivity to tissue microstructure and degeneration. It can be applied across diverse acquisition types and datasets (in vivo and ex vivo), from Cartesian encoding to advanced spiral techniques.