Deep Learning-Based Diagnosis of Major Depressive Disorder Using Electroencephalography Signals

Abstract

Not only is early diagnosis of Depression a critical issue as it is a fast-growing disease in this current decade, but also collecting data from this group of patients is very challenging as they cannot be relaxing and stay still. In this study, we aimed to use a limited number of Electroencephalography (EEG) electrodes to make it easier for the clinician and patient and develop a Conventional Neural Network (CNN) model to help the clinician to have a more accurate diagnosis of the disease. For this aim, we used two public EEG data of people with Major Depressive Disorder (MDD). Totally, the EEG data of 91 subjects (38 healthy and 53 MDD) were utilized in our study. After preprocessing, the spectrogram of EEG was used as input for the CNN model which was Mobile Net V2. The highest validation accuracy we could get was 91 % with a learning rate and batch size of 0.0001 and 32, respectively. Our results demonstrate that Mobile NET may have a high performance in MDD diagnosis. © 2023 IEEE.