Graph-based EEG approach for depression prediction: integrating time-frequency complexity and spatial topology.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Depression is a leading global mental health issue, posing diagnostic challenges due to subjective and varied symptoms.
• Traditional depression diagnosis lacks objectivity, necessitating advanced, reliable identification methods.
• Electroencephalogram (EEG) signals offer a biologically reflective and accessible measure of brain activity, crucial for objective assessment.

Purpose of the Study:

• To develop and validate an objective and effective method for depression prediction using EEG signals.
• To integrate time-frequency complexity and spatial topology of EEG data for enhanced diagnostic accuracy.
• To establish EEG time-frequency complexity as a potential biomarker for depression.

Main Methods:

• Extracted time-frequency complexity and temporal features from EEG signals to create node features for a graph convolutional network (GCN).
• Calculated the brain network adjacency matrix using channel correlation to represent electrode spatial topology.
• Trained and validated a GCN model using the derived node features and adjacency matrix for depression classification.

Main Results:

• Achieved high accuracy rates of 98.30% on the MODMA dataset and 96.51% on the PRED+CT dataset.
• Demonstrated the reliability and utility of the proposed EEG-based depression prediction strategy.
• Confirmed the effectiveness of EEG time-frequency complexity as a valuable biomarker for depression.

Conclusions:

• The proposed GCN strategy effectively merges EEG time-frequency complexity and spatial topology for accurate depression prediction.
• This method offers a promising, objective approach to aid in the clinical diagnosis of depression.
• EEG time-frequency complexity characteristics are validated as significant biomarkers for depression identification.