Epileptic Seizure Prediction in Scalp EEG using One Dimensional Local Binary Pattern based Features

Abstract

Seizure prediction will deeply improve the quality of life of epileptic patients. In this paper, a new method of automatic seizure prediction is presented using one dimensional local binary pattern (1D-LBP) based features in scalp electroencephalogram (EEG). In the feature extraction stage, the preictal and interictal EEG signals were transformed to the 1D-LBP domain and histogram features were extracted. These features were submitted to two different types of classifiers: linear discriminant analysis (LDA) and support vector machine (SVM). In order to reduce the false prediction rate (FPR), a simple post processing stage was also incorporated. The classification using SVM showed improvement over LDA in terms of sensitivity, prediction time and FPR. The proposed method was evaluated using the scalp EEG recording from 13 patients with a total number of 47 seizures. It could achieve a sensitivity of 96.15%, an average prediction time of 51.25 minutes with an FPR of 0.463.