A Gradient-Guided Spatio-Temporal Graph Convolutional Network for Population-Level Major Depressive Disorder...

Area of Science:

• Neuroscience
• Psychiatry
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Major Depressive Disorder (MDD) diagnosis is challenging due to symptom heterogeneity.
• Graph neural network (GNN) based brain network analysis shows promise for MDD diagnosis.
• Existing GNN methods often neglect brain's hierarchical organization and non-imaging data.

Purpose of the Study:

• To develop an advanced GNN model for accurate population-level MDD identification.
• To address limitations in existing brain network construction for MDD diagnosis.
• To integrate spatio-temporal dynamics and non-imaging data for improved MDD modeling.

Main Methods:

• Proposed a gradient-guided spatio-temporal graph convolutional network (GST-GCN).
• GST-GCN models multi-scale temporal dependencies in blood oxygen level dependent (BOLD) signals.
• Employed gradient-guided hierarchical pooling for spatial topology construction and integrated non-imaging data into a subject-level population graph.

Main Results:

• Achieved 91.17% classification accuracy on the REST-meta-MDD dataset, outperforming state-of-the-art methods.
• The GST-GCN framework demonstrated superior performance across all evaluated metrics.
• Identified key biomarker regions and revealed functional gradient alterations associated with MDD.

Conclusions:

• The proposed GST-GCN model offers a robust and interpretable approach for MDD diagnosis.
• Integrating spatio-temporal brain network features and non-imaging data significantly enhances diagnostic accuracy.
• The findings highlight the neurobiological underpinnings of MDD and the potential of AI in psychiatric diagnostics.