earticle

영어

Recently, deep learning-based computer-aided diagnosis (CAD) systems using electroencephalography (EEG) have shown promising potential in the diagnosis of Major Depressive Disorder (MDD). However, most existing models rely on single-branch convolutional neural networks (CNNs), which are limited in their ability to capture spatiotemporal interactions within EEG signals, thereby restricting diagnostic performance. To address this limitation, we propose a residual multi-branch CNN designed to effectively learn both temporal and spatial features from EEG data for the diagnosis MDD patients. To this end, resting-state EEG recordings were collected from 90 drug-naïve MDD patients and 90 matched healthy controls. Frequency-domain features were extracted, and explainable AI techniques (Grad- CAM), along with statistical analysis, were used for informative channel selection. Our model achieved a classification accuracy of 96.11% using all 62 EEG channels and maintained high performance (89.44%) with only 10 channels. These findings demonstrate the potential of our approach for accurate and efficient early-stage MDD diagnosis in clinical applications.