Structure-Aware Adaptive Kernel MPPCA Denoising for Diffusion MRI

TL;DR

This paper introduces an adaptive kernel MPPCA method for diffusion MRI denoising, which dynamically selects patch sizes based on local structures to enhance noise reduction in high b-value DWI images.

Contribution

The proposed ak-MPPCA method adaptively chooses patch sizes for each voxel, improving denoising performance over fixed-size approaches in structurally diverse regions.

Findings

Enhanced denoising quality in high b-value DWI images.

Better preservation of structural details during noise reduction.

Outperforms traditional fixed patch size MPPCA methods.

Abstract

Diffusion-weighted MRI (DWI) at high b-values often suffers from low signal-to-noise ratio (SNR), making image quality poor. Marchenko-Pastur PCA (MPPCA) is a popular method to reduce noise, but it uses a fixed patch size across the whole image, which doesn't work well in regions with different structures. To address this, we propose an adaptive kernel MPPCA (ak-MPPCA) that selects the best patch size for each voxel based on its local neighborhood. This improves denoising performance by better handling structural variations.