A Novel Lbp-Based Algorithm for Automatic Diagnosis of Epileptic Seizures

Abstract

Epilepsy is a condition of brain dysfunction which affects about 1% of the population across the globe. Diagnosing seizures is an unavoidable component in its treatment and control. Epilepsy detection is commonly done using electroencephalogram (EEG) signals. A new EEG based methodology for automatic diagnosis of epileptic seizure has been proposed in the present work. Local Binary Pattern (LBP) values were computed on the preprocessed EEG signal and the morphological significance of LBP values were analyzed, from which eight significant LBP values were selected, whose histogram per each epoch was considered as features. This algorithm was tested for its performance on Children’s Hospital Boston–Massachusetts Institute of Technology (CHB-MIT) EEG database for three different classifiers, namely Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Linear Discriminant Analysis (LDA). Among the three classifiers, KNN shows better performance with 100% Sensitivity and 0.52/h false detection rate (FDR). These values point to the superiority of the present approach over the existing approaches for automatic diagnosis of epilepsy.