Ictal-Interictal Epileptic State Classification with Traditional and Deep Learning Architectures

Abstract

Epileptic seizures are caused by disturbances in the electrical activity of the brain. Failure to correctly classify epileptic forms may result in inappropriate treatment. Activities occurring prior to ictal activity may be causal and require further investigation. Therefore, it is important in the diagnosis of epilepsy to distinguish between ictal and interictal EEG using Electroencephalography (EEG) signs. In this study, ictal (absence seizure) and interictal EEG recordings were scored using 4 bipolar (C3-P3, T5-O1, FP2-F8, C4-T4) channel EEGs from Temple University Hospital (TUH) EEG Seizure Corpus (TUSZ) data. The data were divided into 3-second epochs and various features were obtained from the data. The data in each epoch were filtered using the Discrete Wavelet Transform (DWT) Daubechies-2 wavelet and were 0-32 Hz. range has been studied. Feature selection was made with Correlation based Feature Selection (CFS). The performances of traditional and deep learning classifier algorithms (Support vector machine (SVM), Long Short-term Memory (LSTM)) were compared and the results were discussed. The highest success rate was 96.96% in 84.86 seconds with the LSTM classifier model and 96.36% in 0.05 seconds with the SVM classifier algorithm