A time-frequency graph fusion framework for Major Depressive Disorder diagnosis in multi-site rsfMRI data

Major Depressive Disorder (MDD) poses a significant global health threat, impairing individual functioning and increasing socioeconomic burden. Accurate diagnosis is crucial for improving treatment outcomes. This study proposes Time-Frequency Text-Attributed DeepWalk (TF-TADW), a framework for MDD classification using resting-state functional MRI data. TF-TADW integrates time-frequency dynamics and brain network topology. A key aspect is the adaptive weighting of time-frequency features via an attention mechanism, enabling personalized representation learning to address MDD heterogeneity and mitigate site-specific biases in multi-site datasets. Matrix factorization simultaneously learns network topology and node attributes, creating a comprehensive brain network embedding. Evaluated on REST-meta-MDD and SRPBS-MDD, TF-TADW achieved accuracies of 80.13% and 91.97%, respectively. The attention mechanism also identified key MDD-related brain regions, enhancing interpretability. These results demonstrate TF-TADW's effectiveness and potential for clinical application.