Physics-Guided Multi-View Graph Neural Network for Schizophrenia Classification via Structural-Functional Coupling

TL;DR

This work addresses schizophrenia classification by explicitly modeling the coupling between structural and functional brain connectivity using a physics-guided framework. It maps SC to FC via Kuramoto oscillator dynamics, refines predictions with a U-Net, and fuses SC and generated FC through a multi-view GNN with a joint loss that combinesCCA-based fusion and supervised classification. The approach achieves state-of-the-art performance on the FBIRN dataset while providing interpretable neuroscience insights through explainer analyses, and it reduces reliance on functional data by enabling accurate FC estimation from SC alone. Overall, the method advances data-efficient, interpretable connectomics for neuropsychiatric diagnosis with implications for understanding SC-FC mechanisms in SZ.

Abstract

Clinical studies reveal disruptions in brain structural connectivity (SC) and functional connectivity (FC) in neuropsychiatric disorders such as schizophrenia (SZ). Traditional approaches might rely solely on SC due to limited functional data availability, hindering comprehension of cognitive and behavioral impairments in individuals with SZ by neglecting the intricate SC-FC interrelationship. To tackle the challenge, we propose a novel physics-guided deep learning framework that leverages a neural oscillation model to describe the dynamics of a collection of interconnected neural oscillators, which operate via nerve fibers dispersed across the brain's structure. Our proposed framework utilizes SC to simultaneously generate FC by learning SC-FC coupling from a system dynamics perspective. Additionally, it employs a novel multi-view graph neural network (GNN) with a joint loss to perform correlation-based SC-FC fusion and classification of individuals with SZ. Experiments conducted on a clinical dataset exhibited improved performance, demonstrating the robustness of our proposed approach.

Figures (2)

• Figure 1: Overview of the proposed framework: (a) SCs were provided as input to the Kuramoto model for FCs simulation which were then input to a U-Net to predict more precise FCs. (b) Both SCs and predicted FCs were treated as separate views and fed into GNNs followed by multi-layer perceptrons (MLPs) to learn low dimensional embeddings which were used to compute a joint loss consisting both CCA-based loss and classification loss for optimizing the whole network.
• Figure 2: Visual representation of connections within both empirical and predicted FCs found significant for classifying into SZ and HC. Connections within the same brain network were visually highlighted with distinct colors, while connections across different networks were depicted in gray. The thickness of the edges reflects the strength of their respective connections in the obtained explanation graph.