Model-based denoising in diffusion-weighted imaging using generalized spherical deconvolution.

Area of Science:

• Medical Imaging
• Neuroimaging
• Diffusion MRI

Background:

• Diffusion MRI is crucial for neuroimaging but often has low signal-to-noise ratio, especially at high b-values.
• This limitation hinders accurate analysis of brain microstructure.

Purpose of the Study:

• To propose and evaluate a model-based denoising technique for diffusion MRI data.
• To address the signal-to-noise ratio limitations in diffusion MRI, particularly for high b-value acquisitions.

Main Methods:

• A generalized multi-shell spherical deconvolution model was applied using a Richardson-Lucy algorithm.
• Denoised diffusion-weighted images (DWIs) were reconstructed using the forward model with denoised coefficients.
• The method operates in diffusion space, complementing existing image-based denoising approaches.

Main Results:

• Demonstrated improved image quality in DWIs, neurite orientation dispersion and density imaging, and diffusional kurtosis imaging (DKI).
• Achieved up to 50% error reduction in DKI, enabling high-quality imaging with only 30 DWIs, a fourfold reduction in scan time.
• Showcased consistent performance on pathological data, including white matter lesions in a multiple sclerosis patient.

Conclusions:

• The proposed generalized spherical deconvolution denoising technique significantly enhances diffusion MRI image quality.
• This method holds potential for improving diagnostic accuracy and reducing scan times in clinical neuroimaging.