Silent Signals: Using EEG and AI to Detect Hidden Mental Health Patterns

The insufficient early objective diagnostic tools often result in underdiagnosing of mental disorders. The topic of the work is new: EEG waves are generic biomarkers that help to determine the early signs of mental disease regardless of its type. Two sets of features were elucidated based on the descriptive characteristics of well-known components of ERP (P300 and N200), which were recorded in auditory oddball experiments. These characteristics were tested on two datasets: one included the traumatic brain injury (TBI) patients, and the other (BSNIP) included numerous types of psychiatric illnesses. The five machine learning methods (logistic regression, SVM, decision trees, random forests, and KNN) were applied to test how useful the features formed were in predicting the results. Oversampling processes were used since the data was not evenly distributed, and this makes the review accurate and fair. Results were also promising. In particular, Random Forest and KNN demonstrated high performance. Means of ERP amplitudes were particularly useful at indicating the mental health, yielding a more pure and less variable signal. These results indicate that features derived in EEG may be used as a foundation of cost-effective and affordable general health screening tools that would allow prompt and wide access to care and early diagnosis.