Classifying subclinical depression using EEG spectral and connectivity measures.

Detecting depression on its early stages helps preventing the onset of severe depressive episodes. In this study, we propose an automatic classification pipeline to detect subclinical depression (i.e., dysphoria) through the electroencephalography (EEG) signal. To this aim, we recorded the EEG signals in resting condition from 26 female participants with dysphoria and 38 female controls. The EEG signals were processed to extract several spectral and functional connectivity features to feed a nonlinear Support Vector Machine (SVM) classifier embedded with a Recursive Feature Elimination (RFE) algorithm. Our recognition pipeline obtained a maximum classification accuracy of 83.91% in recognizing dysphoria patients with a combination of connectivity and spectral measures. Moreover, an accuracy of 76.11% was achieved with only the 4 most informative functional connections, suggesting a central role of cortical connectivity in the theta band for early depression recognition. The present study can facilitate the diagnosis of subclinical conditions of depression and may provide reliable indicators of depression for the clinical community.