EEG-based Characterization and Classification of Severity for the Diagnosis of Post-Traumatic Stress Disorder (PTSD)

Post-Traumatic Stress Disorder (PTSD) is a complex syndrome that may occur after overwhelming circumstances frequently involving threatened death or injury that consist of intrusive nightmares, an emotional numbing, a constant state of high arousal, and avoidance of recurring events. However, PTSD severity can vary depending on the traditional methods that the patient conducted. Today, neuropsychiatrists and related professions typically pair electroencephalography (EEG) usage with clinicians’ standard, subjective clinical assessments and patients’ self-reports for diagnostic specificity for PTSD treatment. In this study, we attempted to characterize the significant features and potential biomarkers that can quantify the severity of a patient based on its recorded electrical brain activity on EEG. We trained machine-learning models, specifically Artificial Neural Networks (ANN) and Support Vector Machines (SVM), that can classify the PTSD severity of nine (9) study volunteers using the extracted features from the EEG signals. Results show that the most significant discriminative features were extracted from: (1) the frontocentral site (i.e., FCz electrode); (2) the right frontocentral site (i.e., FC4 electrode); and (3) the left parietal site (i.e., P3 electrode), described by the zero crossing rates, fractal dimensions, and beta-alpha ratios. It can be shown that PTSD severity increases as activity in these regions increases. Furthermore, the trained ANN only required one hidden layer with ten neurons for effective classification, while SVM required a third-order cubic polynomial of its kernel function. After training and testing, results show a testing accuracy of 97.75% for ANN and 99.07% for SVM. Thus, our proposed systems were able to accurately classify the severity of PTSD.