Epileptic Eeg Signal Classification Based On Wavelet Packet Transform And Multivariate Multiscale Entropy

Epilepsy is one of the most common neurological diseases, it disserves people’shealth seriously. The total number of epilepsy is50million worldwide. The EEG ofepilepsy patients appears high amplitude synchronous rhythm wave during seizures, suchas spikes, sharp waves and spike-slow wave est., so electroencephalograph (EEG) is animportant tool for the diagnosis of epilepsy. However, the directly diagnosis of epilepsybased on the patient’s EEG is still difficult because that the EEG of epilepsy patients don’talways show abnormal. Therefore the most important issue in the diagnosis of epilepsy areanalyzing the EEG automatic with computer and extracting epilepsy feature of the EEG..The approaches for analyzing epileptic EEG signals at present are time domainanalysis, frequency domain analysis, time-frequency analysis and non-linear analysis andso on. Wavelet packet transform is one of time-frequency analysis; it has a goodcharacterization of the local signal information in the time domain and frequency domain.However, the classification of epileptic EEG using wavelet packet transform only requireshigh computational complexity, and a lot of storage space. Multivariate multi-scaleentropy is evolved based on sample entropy and multi-scale sample entropy, it not onlycan process multi-channel data parallel, but also can analyze multivariate sample entropyat multi-scale. The main contribution of this paper is:we analyze epileptic EEG usingWavelet packet transform and Multivariate multiscale entropy at the same time. Comparedwith the traditional method proposed method overcomes the shortcoming of traditionalmethod that uses a single method to analyze the EEG signal. Combining two methodstogether, we can avoid the too large data of the compare feature vector and the interferenceof high-frequency signals. So is more conducive to practical application.Firstly, this paper introduces the algorithm principle and application of waveletpacket transform methods and the multivariate multiscale entropy; and then a new methodcombining wavelet packet transform and multivariate multiscale entropy for theclassification of Epilepsy EEG signals is introduced. First the original EEG signals aredecomposed at multi-scales with the wavelet packet transform, and the wavelet packet coefficients of the required frequency bands are extracted。This process filtere off both thecontents of the high-frequency signal and the low-frequency signals of the interferencesignal,which can distinguish the signals better; secomd the wavelet packet coefficients areprocessed with multivariate multiscale entropy algorithm, and we select the better scalerange to classification EEG signals; Finally, the EEG datas are classified by support vectormachines (SVM). The experimental results on the international public Bonn epilepsy EEGdataset show that the proposed method can efficiently extract epileptic features and the accuracyof classification result is satisfactory.