Interpretable deep learning for depression detection in neurological patients using EEG signals.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Depression impacts over 280 million globally, with neurological patients facing higher risks.
• Current depression diagnosis relies on subjective methods, hindering consistent and reproducible clinical assessments.
• Objective diagnostic tools are needed to complement clinical judgment, especially for medication-induced depression in neurological conditions.

Purpose of the Study:

• To develop and validate an interpretable deep learning framework for objective depression detection using electroencephalogram (EEG) signals.
• To combine high-accuracy EEG-based depression classification with explainable AI for clinical transparency.
• To identify specific EEG biomarkers indicative of depression in neurological patients.

Main Methods:

• A lightweight deep learning model was developed and trained on EEG data from 232 neurological patients.
• The model integrated EEG-derived features for depression classification.
• SHAP (SHapley Additive exPlanations) analysis was employed to interpret the model's predictions and identify key biomarkers.

Main Results:

• The deep learning framework achieved 98% classification accuracy in detecting depression.
• SHAP analysis identified significant EEG biomarkers, including the delta/alpha ratio and theta band power.
• The model demonstrated high diagnostic accuracy while maintaining clinical interpretability.

Conclusions:

• The proposed interpretable deep learning framework offers a transparent, accurate, and scalable EEG-based tool for depression detection.
• This objective approach can reduce reliance on subjective evaluations, enhancing consistency in mental health care.
• The framework supports clinical adoption by integrating with existing EEG infrastructure and complementing clinical decision-making.