Multi-view graph network learning framework for identification of major depressive disorder.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) reveals functional connectivity (FC) with non-Euclidean topological structures relevant to major depressive disorder (MDD).
• Existing methods struggle to integrate diverse structural information from FC for comprehensive MDD diagnosis.

Purpose of the Study:

• To develop a novel hierarchical learning structure, the Multi-View Graph Neural Network (MV-GNN), for improved MDD patient identification.
• To integrate both topological and spatial structural information from FC for a more robust diagnostic model.

Main Methods:

• The MV-GNN filters and reconstructs FC from a topological view using various thresholds to identify key brain region attributes.
• A dual leave-one-out cross-feature selection method extracts spatial view information from uniformly sized FC structures.
• Graph convolutional neural networks and self-attention graph pooling are used for efficient embedding, with gating mechanisms fusing topological and spatial views.

Main Results:

• The MV-GNN achieved an average accuracy of 65.61% in identifying MDD patients at a single-center site via leave-one-site cross-validation.
• The model demonstrated superior performance compared to state-of-the-art methods in MDD recognition.
• Differential information between MDD and healthy control (HC) groups was revealed from multiple perspectives.

Conclusions:

• The proposed MV-GNN effectively integrates multi-view FC data for enhanced MDD diagnosis.
• The model offers valuable discriminatory information for the objective diagnosis of MDD.
• This approach serves as a reference for understanding the pathological foundations of MDD.