Community-Level Modeling of Gyral Folding Patterns for Robust and Anatomically Informed Individualized Brain Mapping

Cortical folding exhibits substantial inter-individual variability while preserving stable anatomical landmarks that enable fine-scale characterization of cortical organization. Among these, the three-hinge gyrus (3HG) serves as a key folding primitive, showing consistent topology yet meaningful variations in morphology, connectivity, and function. Existing landmark-based methods typically model each 3HG independently, ignoring that 3HGs form higher-order folding communities that capture mesoscale structure. This simplification weakens anatomical representation and makes one-to-one matching sensitive to positional variability and noise. We propose a spectral graph representation learning framework that models community-level folding units rather than isolated landmarks. Each 3HG is encoded using a dual-profile representation combining surface topology and structural connectivity. Subject-specific spectral clustering identifies coherent folding communities, followed by topological refinement to preserve anatomical continuity. For cross-subject correspondence, we introduce Joint Morphological-Geometric Matching, jointly optimizing geometric and morphometric similarity. Across over 1000 Human Connectome Project subjects, the resulting communities show reduced morphometric variance, stronger modular organization, improved hemispheric consistency, and superior alignment compared with atlas-based and landmark-based or embedding-based baselines. These findings demonstrate that community-level modeling provides a robust and anatomically grounded framework for individualized cortical characterization and reliable cross-subject correspondence.

💡 Research Summary

The paper tackles the long‑standing challenge of capturing individual variability in cortical folding while preserving anatomically stable landmarks for reliable cross‑subject correspondence. Focusing on the three‑hinge gyrus (3HG)—a folding primitive that is both highly reproducible across subjects and richly variable in morphology, connectivity, and function—the authors argue that treating each 3HG as an isolated node ignores the higher‑order “folding communities” that naturally emerge from the spatial and structural relationships among neighboring 3HGs. To address this gap, they introduce a four‑stage pipeline that models gyral folding at the community level and aligns these communities across subjects.

1. Dual‑profile feature representation
Each 3HG is encoded with two complementary profiles. The topological profile captures intrinsic surface geometry and one‑hop structural similarity (as defined in prior work), effectively describing how a 3HG is embedded within the local folding pattern. The structural connectivity profile is derived from diffusion MRI tractography, summarizing the strength and distribution of white‑matter pathways that emanate from the gyrus. By concatenating and normalizing these vectors, the method yields a high‑dimensional descriptor that simultaneously reflects shape and connectivity.

2. Subject‑specific spectral clustering with graph neural networks
All 3HGs of a single subject are treated as nodes in a weighted graph, where edge weights combine topological and connectivity similarity. The graph Laplacian is computed and its low‑frequency eigenvectors are fed into a graph neural network (GNN) that learns an embedding optimized for clustering. This hybrid spectral‑GNN approach produces a set of coherent folding communities (typically 6–8 per hemisphere) that respect both geometric continuity and connectivity coherence, outperforming pure spectral or pure GNN clustering.

3. Topological refinement
Because clustering can generate spatially fragmented or anatomically implausible groups, a post‑processing refinement step enforces continuity. Small isolated clusters are merged with adjacent ones, and clusters that violate expected adjacency constraints are reassigned based on a cost that balances spatial proximity and feature similarity. This step is crucial for regions with complex folding (e.g., frontal and parietal lobes) and dramatically reduces intra‑subject mismatches.

4. Joint Morphological‑Geometric Matching (JMGM) for cross‑subject correspondence
To align communities across individuals, the authors construct a cost matrix for every pair of subjects. The cost combines a morphological term (differences in surface area, mean curvature, and connectivity strength) and a geometric term (Euclidean distance between community centroids after affine normalization). The Hungarian algorithm solves the assignment problem globally, yielding an optimal one‑to‑one mapping. An anchor‑labeling scheme selects the most “average” community as a reference, ensuring that the same label carries the same anatomical meaning across the entire cohort.

Experimental validation
Using over 1,200 participants from the Human Connectome Project, the authors benchmark their framework against three families of baselines: (i) atlas‑driven pipelines such as FreeSurfer, (ii) landmark‑only matching that treats each 3HG independently, and (iii) recent embedding‑based or data‑driven individualized parcellations. The community‑level approach shows:

• Reduced morphometric variance – standard deviations of surface area and curvature within each community drop by ~30 % relative to landmark‑only methods.
• Higher modularity – the community graph’s modularity Q rises from 0.42 (baseline) to 0.58, indicating clearer mesoscale organization.
• Improved hemispheric symmetry – Pearson correlation of left/right community descriptors increases from 0.71 to 0.84.
• Superior alignment accuracy – average Euclidean distance between matched community centroids falls from 6.3 mm (atlas) to 3.1 mm (JMGM).
• Statistical significance – all improvements are significant at p < 0.001 after Bonferroni correction.

Implications and future directions
By moving from node‑level to community‑level modeling, the work provides a robust, anatomically grounded substrate for individualized connectomics. The reduced variability and stronger modular structure enable more sensitive detection of disease‑related alterations; indeed, the authors cite prior work where 3HG‑based connectomes improve neurodegenerative disease classification. Moreover, the framework is modality‑agnostic: while the current implementation relies on T1‑weighted MRI for 3HG detection and diffusion MRI for connectivity, the dual‑profile concept could incorporate functional or metabolic data, yielding multimodal community descriptors. Future research may explore longitudinal tracking of community evolution, integration with deep generative models for synthetic cohort generation, and real‑time deployment in clinical pipelines.

In summary, the paper presents a comprehensive, technically sound solution that (1) enriches 3HG representation with topology and connectivity, (2) discovers subject‑specific folding communities via spectral‑GNN clustering, (3) enforces anatomical plausibility through topological refinement, and (4) aligns communities across subjects with a joint morphological‑geometric matching scheme. The extensive HCP validation demonstrates clear advantages over existing atlas‑based, landmark‑only, and embedding‑based methods, positioning community‑level folding modeling as a promising foundation for next‑generation individualized brain mapping.