Automated tract extraction via atlas based Adaptive Clustering

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Advancements in high angular resolution diffusion-weighted imaging (HARDI) and tractography increase the need for tract-based analyses.
• Understanding brain structural mechanisms relies on statistical analysis of white matter tracts as connectivity pathways.
• Current tract-based studies face challenges in consistent, automated extraction of white matter tracts across individuals without manual region of interest definition.

Purpose of the Study:

• To design and validate a framework for automated extraction of white matter tracts.
• To overcome limitations of manual region of interest selection in tractography.
• To enable large-scale, reliable tract-based brain connectivity studies.

Main Methods:

• Developed a framework with three components: connectivity-based fiber representation, a fiber bundle atlas, and Adaptive Clustering.
• Utilized connectivity signatures for easy fiber correspondence across subjects.
• Employed group-wise clustering to generate a fiber bundle atlas and Adaptive Clustering with the atlas as a prior for automated new subject tract clustering.

Main Results:

• Demonstrated the applicability, reliability, and repeatability of the automated framework on HARDI scans of healthy individuals.
• The framework successfully extracts white matter tracts without manual seed region selection or region of interest drawing.
• Experimental results confirm the robustness of the automated tract extraction process.

Conclusions:

• The proposed framework automates white matter tract extraction, enhancing consistency and comparability in large-scale studies.
• By removing the need for manual region of interest definition, the framework expands potential clinical applications.
• This approach facilitates tract-based analyses on large samples, advancing the understanding of brain connectivity.