DPGR-Net: A disentangled population-guided graph neural network for major depressive disorder identification.
Authors
Affiliations (6)
Affiliations (6)
- College of Medicine and Biological Information Engineering, Northeastern University, Shenyang, China. Electronic address: [email protected].
- College of Medicine and Biological Information Engineering, Northeastern University, Shenyang, China. Electronic address: [email protected].
- College of Medicine and Biological Information Engineering, Northeastern University, Shenyang, China. Electronic address: [email protected].
- College of Medicine and Biological Information Engineering, Northeastern University, Shenyang, China. Electronic address: [email protected].
- College of Medicine and Biological Information Engineering, Northeastern University, Shenyang, China. Electronic address: [email protected].
- School of Computer Science and Engineering, Northeastern University, Shenyang, China. Electronic address: [email protected].
Abstract
Major depressive disorder (MDD) is a brain disorder characterized by substantial individual heterogeneity, and its accurate diagnosis remains challenging in clinical practice. Graph neural networks applied to functional brain networks (FBNs) constructed from resting-state functional magnetic resonance imaging (rs-fMRI) data have shown promise for MDD diagnosis. However, in multicenter settings, site differences are often entangled with disease-related information, which not only interferes with the model's learning of disease patterns but also compromises the reliable modeling of topological patterns in individual FBNs, thereby limiting identification performance. To address this problem, we propose a disentangled population-guided graph neural network with dynamic refinement (DPGR-Net) for MDD identification. Specifically, DPGR-Net first constructs individual FBN from rs-fMRI data and learns discriminative subject-level brain graph representations. To mitigate the interference of site effects, we disentangle disease-related associations from site-related variations at the population level and impose an independence constraint to learn more stable cross-site disease-discriminative representations. On this basis, we innovatively design a population-guided dynamic optimization mechanism for individual brain graphs, which feeds the stable disease-discriminative information learned at the population level back into the structural updating of individual brain graphs, thereby enhancing disease-related connections while suppressing site-related noisy connections. Experimental results on the public REST-meta-MDD dataset show that the proposed method achieves an ACC of 92.37% and outperforms existing state-of-the-art methods. Furthermore, DPGR-Net identifies discriminative brain regions and aberrant functional connectivity patterns associated with MDD, thereby advancing the interpretability and clinical utility of rs-fMRI for MDD diagnosis.