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

Area of Science:

• Medical Imaging
• Image Processing
• Neuroscience

Background:

• Diffusion MRI is crucial for neuroimaging but susceptible to noise, especially with limited diffusion directions.
• Accurate denoising is essential for reliable diffusion tensor imaging (DTI) metrics and tractography.

Purpose of the Study:

• To introduce a novel two-step nonlocal principal component analysis (PCA) method for denoising diffusion MRI.
• To demonstrate its effectiveness in complex diffusion MR images acquired with few diffusion directions.

Main Methods:

• Implemented a two-step denoising pipeline with accurate patch selection and preprocessing (g-factor normalization, phase stabilization).
• Utilized a nonlocal PCA algorithm with a data-driven optimal shrinkage for singular value manipulation to estimate noise-free signals.
• Evaluated performance using simulated and in vivo human data, comparing with local PCA methods.

Main Results:

• Substantially enhanced image quality in both simulated and human data, outperforming the noisy counterpart.
• Improved estimation of diffusion tensor imaging (DTI) metrics and whole-brain tractography.
• Outperformed existing local PCA methods in noise reduction while preserving anatomical details.

Conclusions:

• The proposed two-step nonlocal PCA method effectively denoises diffusion MRI with few diffusion directions.
• This technique enhances image quality, benefiting applications like parametric mapping that use limited diffusion data.