EEGMetaNet: An Enhanced Major Depressive Disorder Detection Framework Using Electroencephalogram and One-Dimensional Convolutional Neural Network

Major depressive disorder (MDD) remains highly challenging to diagnose using objective methods, because current clinical assessments rely heavily on subjective evaluation. Although deep learning (DL) frameworks, which employ electroencephalography (EEG), have demonstrated effective results in detecting MDD, in most existing studies, the focus has been mainly on resting-state EEG signals. As such, task-based EEG data and demographic metadata have frequently been overlooked. Furthermore, existing studies have encountered relevant issues, such as subject-dependent data leakage and overfitting, which can reduce the reliability of the framework. In the present study, EEGMetaNet, a five-layer one-dimensional convolutional neural network (1D-CNN), was proposed by incorporating resting-state and task-based EEG signals with demographic metadata, including age and sex, for automated MDD detection. The proposed framework was evaluated through five-fold Stratified Group K-fold cross-validation to minimise subject leakage while improving the reliability of model evaluation. According to the experimental results, the proposed framework achieved a mean accuracy of 96.55 ± 4.60%. Evaluation on the independent held-out test set yielded an accuracy of 98.68%, a sensitivity of 99.68%, a specificity of 97.17%, an F1-score of 98.90%, and an AUC of 98.54%. Hence, the results suggested that integrating multimodal EEG data with demographic metadata with a subject-wise validation method could assist in improving the DL-based MDD classification. The findings also suggested that the proposed approach could provide a useful computational framework for MDD detection, with further validation in larger and more diverse clinical populations warranted before clinical implementation.