TDSFE-Net: A Temporal Dual-Stream Feature Extraction Network for Depression Detection From EEG

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Psychiatry

Background:

• Early depression detection is crucial for management.
• Electroencephalography (EEG) offers objective diagnostic potential.
• Decoding complex EEG signals for depression is challenging due to non-linearity and dynamism.

Purpose of the Study:

• Introduce a novel Temporal Dual-Stream Feature Extraction Network (TDSFE-Net) for depression diagnosis.
• Enhance the accuracy of depression detection using EEG.
• Investigate the correlation between specific brain region activity and depression.

Main Methods:

• Developed a Temporal Dual-Stream Feature Extraction Network (TDSFE-Net) incorporating attention mechanisms.
• Utilized a hierarchical temporal-separable convolutional network (TSCN) with a local-global attention mechanism.
• Implemented a channel-wise module for spatial dimension analysis and a softmax classifier.

Main Results:

• Achieved high classification accuracies: 98.72% (MODMA), 96.91% (HUSM), and 99.53% (Hospital datasets).
• Demonstrated the network's effectiveness in capturing temporal dynamic characteristics of EEG signals.
• Identified correlations between specific brain region activity and depression.

Conclusions:

• TDSFE-Net shows significant promise for objective, accurate depression diagnosis via EEG.
• The findings provide a scientific basis for discovering depression biomarkers.
• Offers new perspectives on the neural mechanisms underlying depression.