Iterative filtering based automated detection of epileptic seizure EEG signals

Abstract

The analysis of non-stationary characteristics present in electroencephalogram (EEG) signal requires a crucial analysis which can reveal a method for diagnosis of neurological abnormalities, especially epilepsy. This thesis presents a new technique for automated classi cation of epileptic EEG signals into two classes and three classes based on iterative ltering of EEG signals. In this work EEG epochs are decomposed into their intrinsic mode functions (IMFs) using iterative ltering. Amplitude envelope (AE) and instantaneous frequency (IF) functions are extracted from these modes, using discrete separation energy algorithm (DESA). The features are extracted from these IMFs and AE-IF functions. Our feature set includes K-nearest neighbor entropy estimator (KNNE), log energy entropy (LEE), Shannon entropy (SE), Poincar e plot parameters (width, length, and area of plot) of extracted AE functions and IMFs. These features are tested for their discriminative strength, on the basis of their p-values, for classi cation of EEGsignals into seizure, seizure-free, and normal. Our proposed method has given a classi cation accuracy (ACC) of 98% with random forest classi er and 97% with C4.5 decision tree for three class classi cation and ACC of 99.5% for a two class classi cation problem, using mentioned feature set. This proposed methodology has obtained a high ACC in the classi cation of seizure, seizure-free, and normal EEG epochs, which can be useful in more accurate diagnosis of epilepsy.