Epileptic EEG Signal Classification Based On Improved Multivariate Multiscale Entropy

Epilepsy is a kind of nervous system diseases caused by complicated etiology. It isknown that of epilepsy patients with epileptic discharge when they are in seizure periods.That is currently the main basis for the diagnosis of epilepsy disorders. To effectivelydiagnose epilepsy, patients often need to be prolonged EEG monitoring. The huge amountof EEG makes the identification of EEG become an onerous and inefficient work. Thediagnosis results are susceptible to doctor’s subjective factors. So the automaticclassification of epileptic EEG is becoming particularly important.After studying epileptic EEG analysis methods commonly used at home and abroad,this article focuses on the development of improved multivariate multiscale entropyalgorithm. Multivariate multiscale entropy is not only a promotion of multiscale entropyon multivariate signal, but also a reflection of nonlinear dynamic correlation. But thetraditional multivariate multiscale entropy needs large amount of calculation and requiresa lot more time and space when the number of channels increases. The traditionalmultivariate multiscale entropy also can not accurately reflect the correlation betweenvariables. In this paper, the improved multivariate multiscale entropy embeds on allvariables at the same time instead of embedding on a single variable in traditional method,to solve the memory overflow while the number of channels rise, and it is more suitablefor the actual multivariate signal analysis.In this paper, the improved multivariate multiscale entropy and wavelet packetdecomposition method are applied to epileptic EEG classification. Epileptic EEGclassification algorithms can be divided into the time domain, frequency domain,time-frequency domain and non-linear analysis. The time-frequency analysis waveletpacket transform not only reflect the frequency characteristics of the signal but also a goodcharacterization of local signal information. However, the classification of epileptic EEGusing wavelet packet transform only requires high computational complexity, and a lot ofstorage space.The improved multivariate multiscale entropy is not only can processmulti-channel data parallel and analyze multivariate sample entropy at multiscale, but also greatly reduces the complexity and redundancy of the original calculation method. Themain contribution of this paper is: both the improved multivariate multiscale entropy andwavelet packet decomposition are applied to epileptic EEG analyze at the same time. Thefusion method can analysis its statistical characteristics and classification characteristicseffectively. This method not only avoids a large number of space-time consumptioncaused by amount of characteristics data, and eliminates the high-frequency signalinterference in traditions frequency analysis.So the proposed method is more conducive topractical application. The improved multivariate multiscale entropy experimental resultson GAERS rat epilepsy EEG data and Bonn epilepsy EEG data show that this method caneffectively extract EEG features of epilepsy and have good statistical properties andclassification accuracy.