Comparing Study of Nonlinear Model for Epileptic Preictal Prediction

Epilepsy is a group of disorders characterized by recurrent paroxysmal electrical discharges of the cerebral cortex that result in intermitted disturbances of brain function. The damage induced by seizure is severe, so it is significant to predict the preictal state of the epileptic seizure. The aim of this work is to compare and estimate the different nonlinear analysis methods in predicting of epileptic seizure, including approximate entropy, Lempel-Ziv complexity, spectral entropy and C₀ complexity. The features of the epileptic EEG signals were extracted by an integrated nonlinear analysis system developed by LabVIEW. Through the experiments of these nonlinear analysis methods, it is concluded that all of them have a potential application for predicting epileptic seizure (t-test), but the each analysis model has obvious differences. The results indicate that approximate entropy and Lempel-Ziv complexity can distinguish preictal and ictal state with 99% confidence (t-test); spectral entropy achieve 96%, and 97% confidence is achieved by C₀ complexity. Comparing with algorithms complexity, spectral entropy is simpler than the others, which computing speed is shorter.