Toward Multi-Dimensional Depression Assessment: EEG-Based Machine Learning and Neurophysiological Interpretation for...

Area of Science:

• Neuroscience
• Computational Psychiatry

Background:

• Depressive disorder (DD) diagnosis relies on subjective self-reports, leading to potential inaccuracies and delays.
• Objective diagnostic methods are needed to improve precision and clinical utility.

Purpose of the Study:

• To develop an interpretable Electroencephalography (EEG)-based machine learning (ML) framework for diagnosing depressive disorder (DD).
• To assess DD symptom severity and predict cognitive vulnerability using EEG functional connectivity.
• To identify novel EEG biomarkers for clinical decision support.

Main Methods:

• An ML framework integrating optimized functional connectivity features (Coherence, Phase Lag Index, Granger causality) was developed.
• Neighborhood Component Analysis (NCA) was used for feature selection.
• Classification and regression models (KNN, ANN) were employed for diagnosis, severity assessment, and cognitive impairment prediction.

Main Results:

• The model achieved high classification performance (97.66% accuracy) using 21 NCA-selected features with a KNN classifier.
• Accurate severity assessment (r² = 0.89) and cognitive impairment prediction (r² = 0.89) were achieved using an ANN regressor.
• This is the first study to utilize EEG connectivity features for predicting DD severity and cognitive impairment.

Conclusions:

• The proposed EEG-based ML approach offers superior performance in DD classification and severity prediction compared to previous methods, using fewer features.
• Frontal and temporal pathway coherence and PLI values in alpha, beta, and gamma bands are identified as critical biomarkers.
• This research lays the groundwork for objective, clinically actionable decision-support tools in psychiatric care for depressive disorder.