Electroencephalogram (EEG) based automated detection of mental disorders using artificial intelligence processing pipelines

Bipolar disorder, major depressive disorder, and schizophrenia often have overlapping symptoms that lead to frequent misdiagnoses. To address the need for an objective, quantitative and accurate tool for diagnosing mental disorders, we developed an AI-based approach using electroencephalography (EEG) signals. Our study analysed data from Seoul National University, including EEG assessments and medical records of 383 subjects: bipolar disorder (n=67), major depressive disorder (n=199), and schizophrenia (n=117). Our method involved three steps: (1) balancing the dataset with SMOTE up-sampling, (2) extracting key features, and (3) employing machine learning and deep learning models for classification. The combination of Independent Component Analysis, ANOVA F-value, and Gradient Boosting yielded the highest accuracy of 96.67% and minimal misclassifications. These results suggest this approach could significantly improve the correct diagnosis of mental disorders, and it is feasible to quantify the EEG signals to obtain an objective computer-aided diagnosis system.