Complex diffusion-weighted image estimation via matrix recovery under general noise models.

Area of Science:

• Medical Imaging
• Biophysics
• Signal Processing

Background:

• Diffusion MRI is crucial for neuroimaging but suffers from noise and artifacts, particularly in accelerated acquisitions and low signal-to-noise ratio (SNR) scenarios.
• Existing methods often use magnitude-only data and simplified noise models, limiting their effectiveness in complex scenarios.

Purpose of the Study:

• To introduce a novel patch-based singular value shrinkage method for robust diffusion MRI estimation.
• To address challenges in low SNR and accelerated diffusion MRI acquisitions.
• To improve signal recovery and preserve diffusion details.

Main Methods:

• The proposed method operates on complex diffusion MRI data post-sensitivity encoding reconstruction.
• It models noise propagation to determine an asymptotically optimal singular value spectrum for shrinkage.
• Strategies for handling phase inconsistencies and optimizing patch construction are incorporated.

Main Results:

• Quantitative validation on synthetic data, in vivo adult, neonatal, and fetal datasets demonstrated the method's effectiveness.
• The approach successfully generated denoised and debiased diffusion MRI estimates.
• Preservation of spatial and diffusion details was superior compared to magnitude-only methods.

Conclusions:

• The patch-based singular value shrinkage method offers significant improvements for diffusion MRI estimation in challenging conditions.
• It provides a robust framework for enhanced image quality and reliable diffusion parameter estimation.
• The method shows promise for clinical applications requiring high-fidelity diffusion MRI data.