P300 intensities and latencies for major depressive disorder detection

Electroencephalogram (EEG)-based diagnosis of major depressive disorder (MDD) may decrease its chances to be misdiagnosed as a bipolar disorder. In this paper, a machine learning (ML) scheme is presented to automate the diagnose process. It is achieved by discriminating the study participants, i.e., the MDD patients and healthy controls based on the features computed from event-related potential (ERP) data. The ERP features such as the P300 amplitudes and the latencies are computed from the study participants at central locations, i.e, Fz, Cz, and Pz. The ERP features are further used as input to the proposed ML scheme. It is followed by rank-based feature selection involving criteria: t-test, receiver operating characteristics (roc) and wilcoxon. For classification purposes, the logistic regression (LR) classifier is utilized. Finally, the P300 intensities are observed significantly higher in the healthy controls as compared with the MDD patients. In addition, the larger P300 latencies are found in the MDD patients as compared with the healthy controls. Based on the differences of ERP features between the 2 groups, the highest classification accuracy is achieved, i.e., 90.5%. It is concluded that the input features such as the P300 intensities and latencies can discriminate the MDD patients from healthy controls based on a single channel ERP data. In conclusion, the ERP features can be utilized to automate the diagnosis of MDD.