Detection and Classification of Epileptiform Activity in EEG of Rats After Traumatic Brain Injury

Abstract

This paper considers the problem of epileptiform activity recognition in EEG of rats before and after Traumatic Brain Injury (TBI). Recognition and classification was based on expert markup of EEG signals with epileptiform activity - Epileptiform Discharges (ED) and normal sleep activity - Sleep Spindles (SS). Proprietary Event Detection Algorithm (EDA) based on time-frequency analysis of wavelet spectrograms was developed in order to extract valuable events from raw EEG records. Epilepsy Prediction Model (EPM) was based on Power Spectrum Density (PSD) and Frequency features of detected events. Resulted predictors were used in logistic regression model, which estimated probabilities of epileptiform activity in particular EEG event. Validation of proposed model was done by multiple train-test division. It was shown that the accuracy of prediction is around 80%. Proposed algorithms were applied for identification of epileptiform activity in long term EEG records of rats. It was proven that algorithms effectively distinguish epilepsy in rats after TBI in comparison with rats after False Surgery. Proposed approaches can be applied used for Post-Traumatic Epilepsy (PTE) diagnostics in the nearest future.