Detection of major depressive disorder using power spectral densities of electroencephalogram signals and deep learning model

Major Depressive Disorder (MDD) is a mood disorder that affects approximately 185 million people worldwide and can lead to serious and long-term emotional depression and even social isolation. If not diagnosed early, it can lead to serious consequences such as suicide. Traditional diagnostic methods are based on DSM-V criteria, but the complexity of the symptoms makes it significantly difficult to make a rapid and accurate diagnosis. The presence of irregularities and electrophysiological deviations in MDD patients can be determined by recording the electrical activity of the brain with the non-invasive Electroencephalography (EEG) method. In this study, an EEG-based model using Multitaper Spectral Analysis and machine learning methods is proposed to support experts in the early detection of MDD. The EEG dataset consists of EEG signals recorded in two conditions, eyes open and eyes closed, from 27 healthy (C) individuals and 29 MDD patients in the resting state. Using Welch, Periodogram and Multitaper Spectral Analysis methods, 49 feature vectors were extracted from EEG signals and the performances of support vector machine (SVM), random forest (RF), bidirectional long short-term memory (BL-LSTM) and nearest neighbor (KNN) machine learning algorithms were compared. Experimental results show that the model using Multitaper Spectral Analysis and BL-LSTM machine learning algorithm has the highest performance. A promising performance was achieved with 96.55% accuracy, 0.9632 sensitivity, 0.9691 precision, 0.9679 specificity, 0.931 Matthews correlation coefficient (MCC), 0.9662 F1-score, and 0.931 Kappa score values ​​by using Multitaper Spectral Analysis and BL-LSTM algorithm. These results show that EEG based models can be used in the early diagnosis of MDD and can significantly benefit the clinical processes.