Community-level modeling of gyral folding patterns for robust and anatomically informed individualized brain mapping.

Area of Science:

• Neuroscience
• Computational Anatomy
• Brain Imaging Analysis

Background:

• Cortical folding exhibits significant individual variability but contains stable anatomical landmarks like the three-hinge gyrus (3HG).
• Existing methods often analyze 3HGs in isolation, neglecting their organization into higher-order folding communities essential for mesoscale structure.
• This isolated analysis oversimplifies gyral architecture and increases sensitivity to positional variability and noise in landmark matching.

Purpose of the Study:

• To propose a novel spectral graph representation learning framework for modeling community-level folding units, not isolated landmarks.
• To develop a robust method for establishing cross-subject correspondence by aligning community-level representations.
• To enable more reliable, high-resolution, subject-specific analyses of cortical organization.

Main Methods:

• Characterized each 3HG using a dual-profile representation integrating topological surface context and structural connectivity.
• Employed a subject-specific spectral clustering module to identify coherent folding communities, followed by topological refinement.
• Introduced Joint Morphological-Geometric Matching (JMGM) to align community representations by optimizing geometric and morphometric similarity.

Main Results:

• The proposed community-level modeling significantly reduced morphometric variance and enhanced modular organization in folding communities.
• Achieved superior cross-subject alignment and improved hemispheric consistency compared to atlas-based and existing landmark/embedding methods.
• Demonstrated robust, anatomically grounded characterization of individualized cortical organization across over 1000 Human Connectome Project subjects.

Conclusions:

• Community-level modeling of gyral landmarks provides a more robust foundation for individualized cortical characterization than isolated landmark approaches.
• This framework enables more reliable correspondence and high-resolution subject-specific analyses, advancing our understanding of brain organization.
• The spectral graph representation learning and JMGM approach offer a powerful tool for detailed neuroanatomical studies.