Moderate Traumatic Brain Injury Identification from Power Spectral Density of Electroencephalography's Frequency Bands using Support Vector Machine

Traumatic brain injury (TBI) is one of the traumas that requires immediate identification and treatment. Delayed in receiving medical attention can bring severe consequences. Conventional way of identifying TBI are from the computed tomography (CT) or magnetic resonance imaging (MRI). However, CT scan or MRI is hard to be performed to all patients as there were limited resources in the hospitals. As an early screening for TBI, EEG is a suitable substitution. Yet, it is time consuming for doctors or expert to perform EEG analysis manually for each patient. Therefore, this study suggests a machine learning method to identify moderate TBI patients. In this study, two type of features were analyzed. First is the raw EEG signal of alpha, beta, theta and gamma frequency bands. The second type of features are the power spectral density (PSD) of the respective frequency bands. The extracted features were used to train a support vector machine (SVM) separately to investigate the most informative frequency band features for identification of moderate TBI. From the result, it is found that alpha band features give the best identification accuracy which is at 87.5% for the raw features and 83.33 % for the PSD.