Metric Classification of Traumatic Brain Injury Epileptiform Activity from Electroencephalography Data

Abstract

Prediction algorithm of Epilepsy Seizures and Sleep Spindles in electroencephalography (EEG) data is studied in this article. EEG data was measured in rats with Post-Traumatic Epilepsy (PTE) before and after Traumatic Brain Injury (TBI). Experts manually partitioned records into two classes: one, which refers to epileptic activity - Epilepsy Seizures, and second class, which refers to normal behavior of rats - Sleep Spindles (SS). Proposed algorithm was trained and tested on the collected data, which contained EEG features, previously extracted by detection algorithm. Feature importance was evaluated, and logistic regression model was built. Cross validation results were 79% Area Under Curve (AUC) for the best model.