Identification of major depressive disorder based on Triple-GCN model constructed with multimodal elastic network from higher-order brain connectivity.

Major Depressive Disorder (MDD) is the common mental health disease threatening human well-being. Several neuroimaging studies show that analyzing neural connectivity patterns improved diagnostic accuracy, though most approaches overlook node-edge interactions. Our study proposed an integrated approach combining LASSO and Ridge regression algorithms with brain connectivity features. Using both sMRI and fMRI data, we constructed a Multi-feature(gray matter volume/ALFF/Reho) Fusion Elastic Net (MFEN) framework to enhance MDD identification. Furthermore, we improved the Graph Convolutional Neural Network (GCN) algorithm by incorporating a self-attention mechanism and applied a triple Siamese Network to enhance feature extraction. Our proposed method of MDD identification was experimented on 2048 first-episode drug-naive MDD patients and 2562 healthy controls, using rs-fMRI data and sMRI features from the UK Biobank database. Results demonstrated that the extracted features significantly enhanced discriminative capability, establishing the foundation for identifying more reliable biomarkers in MDD patients. By integrating these techniques with elastic networks, the classification accuracy for MDD detection improved substantially to 89%, highlighting the framework's superior performance in mental health diagnostics. In summary, this MDD identification framework proved highly effective and may offer novel insights for auxiliary diagnosis of other neuropsychiatric disorders in clinical practice.