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

Cortical folding shows substantial inter-individual variability yet contains stable anatomical landmarks that can support fine-scale characterization of cortical organization. Among these landmarks, the three-hinge gyrus (3HG) is a particularly informative folding primitive, exhibiting strong intra-species consistency alongside meaningful variations in morphology, connectivity, and functional relevance. However, prior landmark-based approaches typically model each 3HG in isolation, overlooking the fact that 3HGs form higher-order folding communities that capture mesoscale organizational structure. Ignoring this community-level organization oversimplifies gyral architecture and makes one-to-one landmark matching highly sensitive to fine-scale positional variability and noise. We propose a spectral graph representation learning framework that explicitly models community-level folding units rather than isolated landmarks. Each 3HG is characterized using a dual-profile representation integrating its topological surface context and structural connectivity fingerprint. A subject-specific spectral clustering module identifies coherent folding communities, followed by a topological refinement step that ensures anatomical continuity. To establish cross-subject correspondence, we introduce Joint Morphological-Geometric Matching (JMGM), which aligns community-level representations by jointly optimizing geometric and morphometric similarity. Across more than 1000 Human Connectome Project subjects, the resulting folding communities exhibit substantially reduced morphometric variance, stronger modular organization and superior cross-subject alignment, together with improved hemispheric consistency, compared to atlas-based and existing landmark- or embedding-based baselines. These results demonstrate that community-level modeling of gyral landmarks provides a robust, anatomically grounded foundation for individualized cortical characterization, enabling more reliable correspondence and high-resolution subject-specific analyses.