Enhancing Depression Detection from Emotion EEG with Temporal-Spatial-Spectral Representation Learning.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychiatry

Background:

• Major Depressive Disorder (MDD) significantly impacts psychosocial functioning and quality of life.
• Objective diagnostic biomarkers for MDD are crucial but challenging to develop.
• Electroencephalography (EEG) offers insights into brain activity patterns associated with MDD.

Purpose of the Study:

• To leverage deep learning on emotion-evoked EEG data for objective depression detection.
• To bridge the gap between observable depression symptoms and underlying neural signatures.
• To develop and evaluate a novel hybrid deep learning model for classifying depressed patients and healthy controls.

Main Methods:

• Collected EEG data from 33 depressed patients (DPs) and 40 healthy controls (HCs) during exposure to happy, neutral, and sad emotional stimuli.
• Proposed a hybrid Emotion EEG CNN-Transformer model (EMOCT) integrating Convolutional Neural Network (CNN) and Transformer blocks.
• EMOCT was designed to capture temporal, spectral, and spatial features from EEG data for comprehensive brain activity representation.

Main Results:

• The EMOCT model achieved high classification accuracies for DP-HC discrimination across different emotional stimuli.
• Accuracies reached 85.97% for happy, 82.83% for neutral, and 85.25% for sad emotion EEG data.
• EMOCT demonstrated superior performance compared to other evaluated models in classifying depressed individuals.

Conclusions:

• The developed EMOCT model shows significant potential as an effective and objective tool for depression diagnosis.
• This approach could lead to improved clinical assessment and management strategies for MDD.
• The findings underscore the utility of deep learning applied to EEG in identifying neural signatures of depression.