Joint denoising of diffusion-weighted images via structured low-rank patch matrix approximation.

Area of Science:

• Medical Imaging
• Biophysics
• Computational Neuroscience

Background:

• Diffusion-weighted imaging (DWI) is susceptible to noise, which can obscure microstructural details.
• Existing denoising methods may not fully exploit the inherent redundancy in DWI datasets.

Purpose of the Study:

• To develop a joint denoising method for MR DWIs.
• To leverage natural information redundancy via low-rank patch matrix approximation.

Main Methods:

• Exploits nonlocal self-similarity and local anatomical/structural similarity within DWI datasets.
• Searches for similar patches across the entire dataset and structures them into a patch matrix.
• Denoises patch matrices using weighted nuclear norm minimization and back-distributes results.

Main Results:

• Achieved significant noise reduction while preserving structural details in simulated and in vivo datasets.
• Outperformed the Marchenko-Pastur principal component analysis denoising method.
• Resulted in improved Diffusion Tensor Imaging (DTI) parametric maps with less noise and enhanced microstructural detail.

Conclusions:

• The proposed method effectively denoises DWI datasets by utilizing both nonlocal and local similarities.
• Weighted nuclear norm minimization-based low-rank patch matrix denoising is effective for diffusion MRI applications like DTI postprocessing.