An Explainable Assessment for Depression Detection Using Frontal EEG

Depression is one of the most common mental diseases that seriously affect people’s daily life. Finding effective and useful biomarkers to early recognize depression has become an urgent trend. This study purposes to evaluate the complexity of electroencephalogram (EEG) sub-bands and provide an explainable assessment for the multi-classification of depression. At first, Higuchi’s fractal dimension (HFD) and sample entropy (SEn) were extracted from the frontal EEG sub-bands and statistically analyzed. Second, the performance of various classifiers in recognizing three depression statuses was verified. Finally, the features’ importance on the model output was assessed and quantified. The results exhibited increased complexity in depression compared with healthy controls and increased with the deepening of depression. The best classification accuracy of 90.5% in discriminating three groups. Furthermore, the top 10 features with the highest impact are dominated by high-frequency features, both the overall output impact and single category output impact. The results provide an interpretable quantitative assessment for depression identification, which might potentially help in future depressive state evaluation and prediction.