Epilepsy Seizure Detection using Non-linear and DWT-based Features

Abstract

Epilepsy is a neurological disorder causing recurrent seizures in patients. The analysis of electroencephalogram (EEG) signals can lead to the detection of epileptic seizures. But due to immensely non-linear and non-stationary nature of EEG, it is very tedious to inspect and interpret these signals visually. Therefore, methods such as discrete wavelet transform (DWT), and non-linear features are used for automated detection of epilepsy from EEG signals. These methods represent discontinuities, repeated patterns and complexities in EEG signals much effectively and have very high discrimination ability to differentiate EEG segments into normal, interictal and ictal classes. In this work, we formed various feature sets using combination of DWT statistical features and non-linear features such as entropies and fractal dimension. These feature sets were used to train a Support Vector Machine (SVM) classifier and various performance parameters were evaluated. One of the feature sets resulted in an overall accuracy of 99.7% and thereby the proposed technique emerged as an effective automatic seizure monitoring software.