White matter bundle registration and population analysis based on Gaussian processes.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Accurate registration of white matter tracts is crucial for understanding brain connectivity and generating population atlases.
• Existing methods often rely on point-to-point correspondences, which can be sensitive to tract variations and interruptions.

Purpose of the Study:

• To propose a novel method for registering white matter tract bundles from diffusion imaging data.
• To extend this method for generating population atlases of white matter tracts.
• To overcome limitations of traditional registration techniques, such as reliance on point-to-point correspondences.

Main Methods:

• Utilizing a Gaussian process representation for tract density maps.
• Employing an inner product similarity measure derived from the Gaussian process framework.
• Driving a diffeomorphic registration algorithm using this similarity measure, independent of tract parametrization.

Main Results:

• Demonstrated robustness to tract interruptions and reconnections.
• Enabled seamless comparison and combination of white matter tract bundles.
• Successfully estimated dense deformations of white matter using tracts as anatomical landmarks.
• Generated a population atlas of fiber bundles from in-vivo data.

Conclusions:

• The proposed Gaussian process framework offers a robust and flexible approach for white matter tract registration and atlas generation.
• This method avoids common pitfalls associated with point-to-point registration, leading to more reliable anatomical analysis.
• The framework has potential for efficient registration in the Gaussian process parameter space, reducing computational demands.