EEG complexity analysis using enhanced entropy features for depression detection and severity classification

Depression is a psychological disorder characterized by persistent feelings of sadness, and questionnaires are commonly employed for the diagnose this condition and its associated symptoms. The identification of depression is critically important due to its potential to induce significant socioeconomic issues and long-lasting effects on the economy. However, utilizing questionnaires allow room for human error in their findings. This study proposes an objective method for detection depression and its severity using EEG signals. Novel entropy based features, including dispersion entropy (DE), permutation entropy (PE) and their enhanced variants, were extracted to quantify EEG complexity across different brain regions and frequency bands. Following that, the number of dimensions in the feature space was reduced utilizing the minimum-redundancy maximum-relevancy (mRMR) feature selection approach. Ultimately, support vector machine (SVM), logistic regression (LR), and k-nearest neighbor (KNN) were implemented as classifiers, and their performances were assessed through 5-fold cross-validation. The results indicated that the SVM classifier had an average accuracy of 97% in differentiating individuals in the normal and depressed categories. Moreover, SVM classifier was able to grade the severity of depression with a 71% accuracy. These findings indicate that entropy-based EEG analysis can serve as a powerful and non-invasive tool for both diagnosis and severity assessment of depression, particularly reflecting functional alterations in the frontal, temporal, and parietal lobes.