When Brain Networks Travel: Learning Beyond Site

2026-05-07

AI Summary

This study addresses the limited generalizability of cross-site brain network modeling, which is hindered by site-related confounding factors and the assumption of static functional connectivity. To overcome these challenges, the authors propose the CORE framework, which jointly achieves site-confound disentanglement, dynamic modeling of transient pathways, and prior-guided individualized adaptive gating. Specifically, CORE employs site-aware disentanglement to extract a reproducible cross-site connectome scaffold, integrates line graph structures with lightweight temporal descriptors to characterize dynamic functional connectivity, and fuses population-level priors with individual-specific features to enhance robustness. Evaluated on four large-scale datasets—ABIDE, REST-meta-MDD, SRPBS, and ABCD—CORE demonstrates an average relative performance improvement of 6.7% and exhibits consistent robustness across diverse brain parcellation schemes, significantly outperforming existing methods.

Abstract

Graph-based learning on functional magnetic resonance imaging (fMRI) has shown strong potential for brain network analysis. However, existing methods degrade under cross-site out-of-distribution (OOD) settings because site-conditioned confounders induce non-pathological shortcuts, while functional connectivity constructed by temporal averaging obscures transient neurodynamics, limiting generalization to unseen sites. In this paper, we propose Cross-site OOD Robust brain nEtwork (CORE), a unified framework for brain network learning across unseen sites. CORE first performs site-aware confounder decoupling to mitigate site-conditioned bias and extract a cross-site population scaffold of reproducible diagnostic connectivity edges. It then profiles transient pathway dynamics over this scaffold using lightweight temporal descriptors and organizes scaffold edges into a line graph for transferable pathway-level modeling. Finally, CORE introduces a prior-guided subject-adaptive gating mechanism that leverages scaffold-derived population priors while preserving subject-specific connectivity variability. Extensive experiments under leave-one-site-out evaluation on real-world datasets (ABIDE, REST-meta-MDD, SRPBS, and ABCD) show that CORE consistently outperforms state-of-the-art baselines, with up to 6.7% relative gain. Furthermore, CORE remains robust to atlas variations, maintaining performance gains across different brain parcellation schemes.

Problem

• cross-site
• out-of-distribution
• functional connectivity
• confounder
• neurodynamics

Innovation

• cross-site generalization
• confounder decoupling
• transient neurodynamics
• line graph modeling
• subject-adaptive gating