Hierarchical Multiscale Structure-Function Coupling for Brain Connectome Integration.

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Integrating structural connectivity (SC) and functional connectivity (FC) in the brain is complex due to non-linear relationships and hierarchical organization.
• Existing methods struggle to capture the multiscale and nested modular nature of brain connectomes.

Purpose of the Study:

• To develop a novel hierarchical multiscale framework for integrating SC and FC.
• To jointly learn individualized modular organization and hierarchical coupling between brain structure and function.

Main Methods:

• Proposed a framework with three key components: Prototype-based Modular Pooling (PMPool) for modality-specific community detection, an Attention-based Hierarchical Coupling Module (AHCM) for SC-FC interaction modeling, and Coupling-guided Clustering loss (CgC-Loss) for cross-modal alignment.
• Utilized individualized modular organization and hierarchical coupling across SC and FC.

Main Results:

• The framework demonstrated superior performance in predicting brain age, cognitive scores, and disease classification across four independent cohorts.
• Consistently outperformed baseline and state-of-the-art approaches on prediction tasks.
• Ablation studies confirmed the effectiveness of individual components.

Conclusions:

• The proposed framework successfully integrates structural and functional brain connectomes at multiple scales.
• Learned coupling visualizations revealed biologically meaningful structure-function relationships, offering interpretable insights.
• The approach advances connectome integration and its application in understanding brain health and disease.