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

TL;DR

The paper tackles cross-subject cortical correspondence by moving from single landmarks to community-level gyral folding representations, centered on stable 3HG landmarks. It develops a spectral graph framework with dual-profile 3HG features, subject-specific GNN clustering with topology-aware refinements, and a Joint Morphological-Geometric Matching strategy to align gyral communities across subjects. Evaluated on over 1,000 HCP subjects, the approach yields lower morphometric variance within communities, stronger modular organization, improved hemispheric consistency, and superior cross-subject correspondence relative to atlas- and landmark-based baselines. This framework advances anatomically informed, robust individualized brain mapping and supports high-resolution analyses of cortical structure and connectivity across large cohorts.

Abstract

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.

Figures (7)

• Figure 1: Overview of current major paradigms for brain mapping and the proposed folding-community-based individualized framework. (a) Atlas-based mapping assigns predefined regional labels via template-driven registration, enabling group-level comparison but often obscuring fine-scale folding variability. (b) Data-driven individualized parcellation and prior landmark-based approaches aim to capture subject-specific patterns. However, data-driven parcellations often lack stable anatomical anchors for establishing cross-subject correspondence, while existing landmark-based methods overlook community-level folding organization and thus remain vulnerable to high inter-individual variability. (c) Proposed folding community-based individualized mapping. Gyral landmarks are identified in each subject, organized into folding communities via graph learning, and aligned across subjects at the community level, yielding anatomically grounded and correspondence-ready folding representations.
• Figure 2: Overview of the proposed pipeline for individualized cortical landmark community-level correspondence. 3HGs are identified on cortical surface, then clustered via subject-specific GNNs with topological refinement to ensure spatial continuity. Cross-subject correspondence of 3HGs communities is obtained through joint morphological–geometric matching, solved with a Hungarian assignment and anchored for consistent cohort-wise labeling.
• Figure 3: Histograms of clustering time for the proposed method using the Destrieux (left) and DK (right) atlases, shown separately for the left hemisphere, right hemisphere, and their mean.
• Figure 4: Boxplots show the within region variance of cortical features for the proposed GyralNet clusters, the Destrieux atlas and the DK atlas after z-score normalization. GyralNet subnetworks consistently display lower variance across curvature, sulcal depth, thickness and area. This indicates that the proposed framework produces more homogeneous regions compared with traditional anatomical atlases. Significance marks ($^{***}$) indicate statistical differences ($p<0.001$) based on the Mann-Whitney U-test.
• Figure 5: Qualitative visualization of cross-subject correspondences established by the proposed JMGM using the Destrieux atlas. Three representative clusters (Cluster-1, Cluster-2, and Cluster-3) are shown across ten subjects, where consistent gyral patterns are highlighted in red, green, and purple, respectively. The bottom panel displays the full set of identified 3HG clusters for each subject, demonstrating stable and anatomically coherent correspondences across individuals.