A localized Richardson-Lucy algorithm for fiber orientation estimation in high angular resolution diffusion imaging.

Area of Science:

• Neuroimaging
• Diffusion MRI
• White Matter Tractography

Background:

• Diffusion tensor imaging (DTI) models white matter fiber trajectories but struggles with crossing fibers.
• High angular resolution diffusion imaging (HARDI) provides richer data but is susceptible to noise and Rician bias.
• Existing spherical deconvolution methods, like Richardson-Lucy (RL), have limitations with Rician distributed HARDI data.

Purpose of the Study:

• To develop a novel spherical deconvolution method for improved accuracy in resolving crossing fibers.
• To address the limitations of conventional methods when analyzing noisy HARDI data.
• To introduce an algorithm that simultaneously reduces noise and corrects Rician bias in diffusion MRI.

Main Methods:

• Proposed a localized Richardson-Lucy (LRL) algorithm for HARDI data analysis.
• The LRL algorithm simultaneously accounts for Rician bias and neighbor correlation in the data.
• Estimated fiber orientations by applying the LRL algorithm to HARDI signals.

Main Results:

• The LRL algorithm demonstrated superior performance in reducing mean angular error (MAE) compared to conventional RL across various noise levels.
• Normalized mean squared error (NMSE) was minimized by the LRL approach, particularly at b-values of 3000 s/mm(2).
• Analysis of real HARDI data showed that LRL accurately depicted fiber structures in complex crossing regions.

Conclusions:

• The novel LRL algorithm effectively reduces noise and corrects Rician bias in HARDI data.
• This method enhances the accuracy of fiber orientation estimation, particularly in areas with complex fiber architectures.
• Experimental validation with synthetic and real data confirms the LRL algorithm's success and effectiveness for white matter tractography.