Leveraging Deep Learning for Enhanced Classification of Depres‎sion Via EEG-Derived Imagery

Depression is the most prevalent psychological disorder worldwide, affecting individuals irrespective of age and frequently associated with ‎underlying organic etiologies. Its influence extends beyond psychological health, exerting significant effects on physical well-being as well. ‎Clinically, depression is recognized as a neuropsychiatric condition linked to alterations in the brain’s neurochemistry, with its pathogenesis ‎involving a complex interaction among biological, genetic, psychological, and environmental determinants. In this study, we developed a ‎deep learning-based approach for the classification of depression using electroencephalogram (EEG)-derived imagery. Specifically, the ‎ResNet50V2 convolutional neural network architecture was employed to differentiate between EEG images of healthy controls and those ‎diagnosed with Major Depressive Disorder (MDD) based on standard diagnostic criteria. A meticulously curated dataset comprising pre-‎processed EEG images from both cohorts was used to train the model. Transfer learning was applied by leveraging the pretrained ResNet50V2 weights from the ImageNet dataset, facilitating efficient feature extraction tailored for the EEG domain. The model’s performance ‎was evaluated using multiple quantitative metrics, achieving a classification accuracy of 97.25%, indicating high discriminative capability. ‎These findings demonstrate the potential of deep learning models, particularly ResNet50V2 with transfer learning, for the reliable detection ‎of depression from EEG images, which may support timely diagnosis and intervention in clinical settings‎.