Explainable Temporal Deep Learning for EEG-Based Depression Detection Using Resting-State Brain Dynamics

Area of Science:

• Computational psychiatry
• Neuroscience
• Artificial Intelligence

Background:

• Depression is a significant mental health disorder.
• Electroencephalogram (EEG)-based automated detection offers a potential objective diagnostic tool.
• Challenges exist in achieving high accuracy and interpretability in EEG signal analysis due to complex spatiotemporal structures.

Purpose of the Study:

• To propose an explainable deep learning framework for depression detection using resting-state EEG data.
• To enhance the accuracy and interpretability of automated depression diagnosis.
• To leverage explainable AI techniques for understanding EEG-based depression classification.

Main Methods:

• A retrospective computational study utilized deep learning on EEG data from 106 subjects (controls and depressive groups).
• A Convolutional Neural Network-BiLSTM architecture with an attention mechanism was developed.
• Explainable AI techniques, including Grad-CAM and SHAP, were integrated for interpretability.

Main Results:

• The model achieved high performance with 89.76% accuracy, 89.58% F1-score, and 0.936 AUC.
• Ablation analysis validated the importance of temporal modeling and attention mechanisms.
• Explainability analysis highlighted the significant role of frontal EEG channels in classification.

Conclusions:

• The developed framework offers an accurate and interpretable method for EEG-based depression detection.
• This approach supports advancements in computational psychiatry and clinical decision-support systems.
• The findings underscore the potential of explainable AI in analyzing complex neurological data for mental health applications.