Machine Learning Based Diagnosis of Post-traumatic Stress Disorder Using Resting-state Quantitative Electroencephalogram

Abstract

Post-traumatic stress disorder (PTSD) is a mental disorder which can occur in people who were exposed to a traumatic event. Recently, the unbiased, objective, and automated methods for the diagnosis of PTSD have attracted increasing attention. However, previous studies on machine-learning-based diagnosis of PTSD with resting-state electroencephalogram (EEG) have reported poor accuracies of as low as 60%. The purpose of this dissertation is to develop a method that can distinguish PTSD patients from the healthy controls (HCs) using only quantitative EEG (qEEG). In the first study, the author investigated the EEG-based conventional classifiers including linear discriminant analysis (LDA), support vector machine (SVM), and random forest (RF) classifiers for classifying PTSD patients and HCs. Eyes-closed resting-state EEG data of 39 HCs and 42 PTSD patients were used for the analysis. EEG source activities in 148 cortical regions were parcellated based on the Destrieux atlas. LDA, radial basis kernel SVM, linear SVM, and RF classifiers employing sensor/source-level band powers and network features as feature candidates were tested. The highest accuracy of LDA, radial basis kernel SVM, linear SVM, and RF classifier has been reported when using the source-level features. The accuracies were 64.14 ± 1.43%, 65.62 ± 3.46%, 61.23 ± 4.17%, and 61.54 ± 3.33%, respectively. In the second study, a Riemannian geometry-based Fisher geodesic minimum distance to the mean (FgMDM) and Riemannian geometry-based logistic regression classifier (Tangent space logistic regression; TSLR) is proposed for PTSD classification for the first time. The identical EEG raw data and source activities which used in first study were also applied in the analysis. Thirty epochs of preprocessed EEG were employed to evaluate their covariances for each individual. The FgMDM classifier showed an average classification accuracy of 73.09 ± 2.08%. The TSLR showed average classification accuracy of 71.36 ± 2.08%. Both classifiers showed improved accuracy when source covariance was applied. In the present study, the author demonstrated that the diagnostic accuracy of PTSD with resting-state EEG could be significantly improved by employing the FgMDM framework, which is a type of Riemannian geometry-based classifier.