Denoising complex-valued diffusion MR images using a two-step non-local principal component analysis approach.

Area of Science:

• Medical Imaging
• Neuroscience
• Biomedical Engineering

Background:

• Diffusion Tensor Imaging (DTI) is crucial for neuroimaging but susceptible to noise.
• Acquiring high-quality DTI data often requires numerous diffusion directions, increasing scan time.
• Effective denoising methods are essential for accurate DTI analysis and clinical applications.

Purpose of the Study:

• To propose and validate a novel two-step non-local principal component analysis (PCA) method for DTI denoising.
• To demonstrate the method's efficacy in improving image quality with a reduced number of diffusion directions.
• To enhance the utility of DTI for applications requiring parametric mapping from limited data.

Main Methods:

• Implemented a two-step denoising pipeline with accurate patch selection for high noise levels.
• Incorporated g-factor normalization and phase stabilization for robust preprocessing.
• Utilized a non-local PCA algorithm with optimal shrinkage for noise-free signal estimation.

Main Results:

• Substantially enhanced DTI image quality in both simulations and human data.
• Outperformed existing local-PCA-based methods in noise reduction while preserving anatomical details.
• Achieved improved estimation of DTI metrics and whole-brain tractography compared to noisy data.

Conclusions:

• The proposed non-local PCA denoising method effectively improves DTI image quality using fewer diffusion directions.
• This approach is beneficial for applications prioritizing parametric mapping with limited imaging volumes.
• The method holds significant promise for advancing DTI-based neuroimaging research and diagnostics.