A Deep Learning Approach for MDD Classification Using Frontal EEG Spectral Features with DenseNet-121

Major Depression Disorder (MDD) is a common yet destructive mental disorder that affects millions of people worldwide. Making early and accurate detection of MDD is essential for timely interventions and effective management. Electroencephalography (EEG) is a valuable tool for exploring the brain-based aspects of depression, offering considerable promise in its diagnosis. This research presents an advanced EEG-based methodology for detecting depressive disorders, utilizing the publicly accessible Multi-Modal Open Dataset (MODMA). EEG data were collected from 53 participants during resting states, employing 25 frontal electrodes. We analyzed five minutes of resting-state EEG data to compute spectrograms for each channel using a 10 -second window with 50% overlap, resulting in a total of 39,750 spectrograms. Various deep learning models were evaluated, with the DenseNet-121 model achieving the highest classification performance, yielding a training accuracy of 98.99% and a validation accuracy of 92.82%. This innovative approach enhances existing methods for classifying major depressive disorder by optimizing both the number of electrodes and classification accuracy. As a result, it holds significant potential for widespread use in psychiatric settings, offering valuable support to psychiatrists in diagnosing and managing MDD.