Hierarchical Multiscale Structure-Function Coupling for Brain Connectome Integration
A novel hierarchical model outperforms state-of-the-art methods on brain age, cognition, and disease prediction tasks.
A research team led by Jianwei Chen has introduced a novel AI framework designed to solve a major challenge in neuroscience: integrating structural and functional brain connectomes. The relationship between the brain's physical wiring (structural connectivity, or SC) and its dynamic activity patterns (functional connectivity, or FC) is complex, non-linear, and organized across multiple scales. The proposed 'hierarchical multiscale structure-function coupling framework' tackles this by jointly learning individualized brain modules and how SC and FC interact across these nested hierarchies.
The framework consists of three key technical components. First, Prototype-based Modular Pooling (PMPool) learns multiscale brain communities by identifying prototypical brain regions. Second, an Attention-based Hierarchical Coupling Module (AHCM) models interactions both within and across different hierarchical levels of SC and FC. Third, a Coupling-guided Clustering loss (CgC-Loss) uses these interaction signals to align communities across the two modalities. This architecture allows the model to capture richer, more nuanced relationships than previous methods.
In rigorous evaluations across four independent cohorts, the model demonstrated superior performance in three critical predictive tasks: estimating brain age, predicting cognitive scores, and classifying neurological diseases. It consistently outperformed existing baseline and state-of-the-art approaches. Importantly, visualizations of the learned 'coupling' revealed interpretable patterns that differed across conditions, suggesting the model is capturing biologically meaningful signals rather than just statistical correlations. This represents a significant step toward more accurate and explainable computational models of the brain.
- The framework integrates structural (SC) and functional (FC) brain connectomes using a novel hierarchical, multi-scale approach.
- It outperformed all baseline models across four cohorts for brain age prediction, cognitive score estimation, and disease classification.
- The model's components, like PMPool and AHCM, produce interpretable visualizations of brain region couplings, revealing biologically meaningful patterns.
Why It Matters
This provides a more accurate and interpretable AI tool for neuroscience research and potential clinical applications in brain health assessment.