Defining the Optimal Alpha Band Connectivity Pathways To Detect Depression Using Explainable Ai

Abstract

Depression is a prevalent mental disorder that affects the mood of patients and is generally diagnosed by paper-based questionnaires. Nowadays combining machine learning and Electroencephalogram (EEG) are more popular to diagnose depression. This study proposes a novel method that only focuses on the connectivity of the Alpha band. 22 Depression patients and 25 healthy control subjects were attended in EEG recording in eyes closed and eyes open condition. After the pre-processing step, the phase lag index (PLI) values between EEG channels were calculated. The Extreme Gradient Boosting (XGB) classifier was used to detect depression. The maximum performance with a 95.22%+/- 1.76% accuracy rate, 94.09%+/- 2.15 recall, and 96.08%+/- 2.54% specificity rate was reached when only eyes-closed values were used as inputs of the classifier. Performance of the classifier for ten selected pathways using explainability analysis of features with the Shapley Additive explanation (SHAP) method decreased to 93.12%+/- 1.73% accuracy rate, 92.92%+/- 1.4% recall, and 94.57%+/- 2.78% specificity rate.