A Deep Learning Method Based on Multi-Modality EEG for Automatic Depression Screening

Abstract

Signal variability is linked to irregularities in time series caused by intrinsic nonlinearities of the neural system and can be measured on variable temporal scales over short time series. By measuring refined complex multiscale permutation entropy (RCMPE) from resting-state electroencephalography (EEG) data, we investigated the presence of a specific range of time scales characterizing major depressive disorder (MDD).We used an EEG dataset acquired from 22 MDD patients and 22 healthy controls in the eyes-closed (EC) and eyes-open (EO) states available on the PRED + CT website. Signal variability in both the EC and EO states was compared between the two groups, and their relationship to depressive symptom severity was examined.In the EC state, the RCMPE was higher in the MDD group than in the control group on a coarse temporal scale, approximately 20–32 ms, at almost all sensors. It also showed a negative correlation with depressive symptom severity on a fine temporal scale, approximately 2–26 ms, in the frontal, right temporal, and left parietal sensor areas in MDD. The EO state revealed a group difference but no relationship with depressive symptom severity.Our results suggested that the diagnosis of MDD as a trait and the severity of depressive symptoms as a state are linked to EEG signal variability on the coarse temporal scale and the fine scale in the resting state, respectively.Signal variability reflects different characteristics of depression depending on the temporal scale.