Research On Synchronization Analysis Of EEG And Epileptic Seizure Prediction

EEG signal is the expression of human complex electrophysiology process,and it contains abundant brain activity information which is related to physiological structure and state of the human brain.Synchronization of EEG signal is a key feature which shows different regions of the brain establishing a correspondence,and it reveals how the brain regions integrate and communicate the information.Synchronization of EEG signal help to understand deeply the mechanism of brain dysfunction and it has very great significance to diagnosis,prevention and treatment the nerve dysfunction disease.Epilepsy is paroxysmal abnormal electrical activity of brain neurons in the clinical manifestations;it is sudden,repeatability and temporary.Epileptic disease brings physical and psychological harm to patients,and seriously even life-threatening.If we can predict the epileptic seizure as soon as possible,it can provide enough time for patients and doctors to take the corresponding measures to prevent it,in order to avoid bringing accident harm to patients.This paper uses the synchronization analysis theory of EEG signals to predict epileptic seizure,deeply study that all kinds of EEG synchronization parameters when epileptic patients in different state,and presents several new methods of epileptic seizure prediction.The specific research content as follows:(1)This dissertation has analyzed the scalp epileptic EEG data from hospital 463 of PLA and the public database CHB-MIT.According to the results of previous studies,synchronization analysis of epileptic EEG widely used cortex EEG signals,which is invasive collection and can't meet the real-time tracking of the patient condition.This paper used the scalp EEG signals,and it is noninvasive collection,which is helpful for the clinical application of epileptic prediction.(2)This dissertation proposed a method of phase synchronization brain network,which analyzed the synchronization of epileptic EEG from the whole brain.First,EEG was divided into interictal period team,preictal period team and ictal period team;frequency band was extracted by using wavelet transform and then the instantaneous phase was calculated by using Hilbert Huang transform;phase locking value was calculated respectively between each lead,constructed functional brain network,and analyzed the parameters of brain network.The parameters in three teams were compared,and the change rule of phase synchronization was found during the seizure process in the whole area of the brain.(3)This dissertation analyzed the synchronization of epileptic EEG signal from the perspective of phase synchronization,spectrum synchronization and frequency synchronization.Based on the mentioned contents,the instantaneous phase,Hilbert Marginal Spectrum and Hilbert weighted frequency were calculated by using Hilbert-Huang transform;the multivariate phase synchronization index was obtained by Multivariate phase synchronization analysis.Through comparing the electroencephalotopogram of multivariate phase synchronization index,Hilbert marginal spectrum and Hilbert weighted frequency,the change rule of the three characteristics was found during the seizure process in the whole area of the brain.(4)This dissertation proposed a new method of epileptic seizure prediction based on spatial-frequency domain feature analysis.Based on previous analysis results,spatial-frequency domain feature characteristics consisted of multivariate phase synchronization index,Hilbert marginal spectrum and Hilbert weighted frequency,and epileptic seizure was predicted by putting the characteristics into classifiers.In order to find the optimal predictive effect,this paper used separately the Fisher linear discriminant analysis,SVM and Extreme Learning Machine to predict the epileptic seizure.Predictive effect was evaluated by K cross validation method and the results showed that the SVM and Extreme Learning Machine had the best effect.Several relative predictive approaches were evaluated by seizure prediction characteristic method,and the results showed that the method based on spatial-frequency domain feature analysis could get higher sensitivity and lower error rate of forecast(the mean false alarm rate:0.05/h;the mean sensitivity:87.92%;the mean forecasting time:37.8 min).(5)This paper proposed the prediction method of epileptic seizure based on phase synchronization factor analysis model.Based on previous analysis results,the factor analysis model of instantaneous phase was calculated,that is say phase synchronization related source corresponding time series.Phase synchronization factor analysis model as the characteristic parameters was put into classifier to predict the epileptic seizure.Using described above three kinds of classifier,the forecast effect were compared.Phase synchronization factor analysis model and principal component analysis model were evaluated by seizure prediction characteristic method,and the results showed that the method based on phase synchronization factor analysis model could make a better prediction(the mean false alarm rate:0.05/h;the mean sensitivity:90%;the mean forecasting time:39.8 min).