Exploring Adaptive Graph Topologies and Temporal Graph Networks for EEG-Based Depression Detection.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Graph Neural Networks (GNNs) show promise for EEG-based depression detection.
• Existing GNN methods often use static graph structures, neglecting individual brain network differences and temporal dynamics.

Purpose of the Study:

• To develop an advanced deep learning algorithm for improved EEG-based depression detection.
• To address limitations of current GNNs by incorporating adaptive graph topologies and temporal information.

Main Methods:

• Proposed an Adaptive Graph Topology Generation (AGTG) module for real-time brain network connectivity modeling.
• Introduced a Graph Convolutional Gated Recurrent Unit (GCGRU) module to capture temporal brain network dynamics.
• Utilized a Graph Topology-based Max-Pooling (GTMP) module for accurate feature extraction.

Main Results:

• The proposed model achieved the highest Area Under the Receiver Operating Characteristic Curve (AUROC) of 83% and 99% on two datasets.
• Comparative analysis demonstrated superior performance against advanced algorithms.
• Validation experiments confirmed the method's effectiveness and advantages.

Conclusions:

• The developed deep learning algorithm effectively detects depression from EEG data by capturing individual-specific and dynamic brain network characteristics.
• The findings offer insights into brain network differences between healthy and depressed individuals.