An Interpretable Functional-Dynamic Synaptic Graph Neural Network for Major Depressive Disorder Diagnosis from rs-fMRI.
Authors
Affiliations (5)
Affiliations (5)
- The Clinical Hospital of Chengdu Brain Science Institute, Sichuan Institute for Brain Science and Brain-Inspired Intelligence, China-Cuba Belt and Road Joint Laboratory on Neurotechnology and Brain-Apparatus Communication, School of Life Science and Technology, University of Electronic Science and Technology of China, Chengdu 610054, P. R. China.
- Department of Radiology, Sichuan Provincial People's Hospital, University of Electronic Science and Technology of China, Chengdu 610072, P. R. China.
- Department of Radiology, Guangzhou First People's Hospital, School of Medicine, South China University of Technology, Guangzhou 510180, P. R. China.
- Information and Data Center, Guangzhou First People's Hospital, School of Medicine, South China University of Technology, Guangzhou 510180, P. R. China.
- Information and Data Center, The Second Affiliated Hospital, School of Medicine, South China University of Technology, Guangzhou 510180, P. R. China.
Abstract
Major depressive disorder (MDD) is a serious, complex psychiatric condition that affects millions of people worldwide. Early diagnosis and biomarker identification are critical for personalized treatment and effective disease monitoring. While resting-state functional magnetic resonance imaging (rs-fMRI) combined with deep learning has facilitated MDD prediction, existing methods often overlook the dynamic temporal characteristics of blood oxygen level-dependent (BOLD) signals and ignore the strength of inter-regional connections, resulting in brain region updates devoid of biological specificity. To this end, a functional-dynamic synaptic graph neural network (FDSyn-GNN) is proposed, which integrates a bidirectional gated recurrent unit (Bi-GRU) timestamp encoding (BGTE) module for modeling dynamic BOLD signals and a synaptic graph Transformer (SGT) module for connection-aware brain region updates. FDSyn-GNN is validated on two large-scale MDD datasets collected across multiple sites, where it outperforms 12 state-of-the-art (SOTA) baseline methods. In addition, extensive ablation and interpretability analyses highlight its potential for biomarker discovery, offering insights into the neural mechanisms underlying MDD. The code is publicly available at https://github.com/ZHChen-294/FDSyn-GNN.