Study On Epileptic EEG Space-time Mechanism Based On Phase Transition

Epilepsy is a kind of nervous system disease with sudden and recurrence,not only causes great harm to the health of patients,but also puts a huge burden to family and society.Electroencephalogram(EEG)shows a larger volatility and synchronism during seizures,which is similar to fluctuations presented in time and space when phase change occurs.Based on above,in this thesis,the theory of phase change was integrated into the research of epileptic EEG signals,characteristics of synchronism of EEG and the temporal-spatial evolution were explored from the perspective of phase transition,to capture the early warning information of occurrence and development of epilepsy,main steps were as follows:1.Epileptic EEG signals was denoising preprocessed with the wavelet analysis method.4 methods were used to threshold process coefficients that decomposed by the wavelet,a comparative analysis of denoising results,it was found that there was a better denoising effect on reconstructed signals that coefficients processed with soft threshold.2.Correlation analysis was conducted on the denoising signals,the correlation coefficients between the channels were calculated,and the correlation coefficients as input of classifier,various methods were used to classify signals,such as support vector machine(SVM),nearest neighbor algorithm,Naive Bayes classifier and so on.Correlation analysis showed that there was a growing tendency of the correlation coefficients of Epileptic EEG signals between channels,which from interval to early stage then to the stage of attack.Classification results indicated that classification effect of SVM was better than other classifiers,the best classification results could be obtained if data segment was equal to 30 s.3.Concepts of phase change idea and Z-score were introduced,the critical threshold was chose by experiments and characteristics of space-time evolution of the signals was studied.Through the experiments discovered that when Z was equal to two,there were some characteristics of the signals to show out.4.The characteristics of space and time was conducted to further verification.Epileptic EEG signals of five dogs with a total of 40 hours were further verified.The results showed that the matching rate of space-time signal characteristics that researched by this method reached 80%,among of them,the accuracy of epilepsy data for certain dogs could reach 91.7%.The the accuracy of the classification based on relativityproposed in this thesis was 93.75%,which could attain a better classification effect on signals.From the phase transition perspective,epileptic EEG signals were studied in this thesis.Based on correlation analysis,combined with though of phase change,the temporal and spatial characteristics of EEG signals before epileptic seizures were studied.The results indicated that as time goes on,the closer to the seizure of epilepsy,the higher the degree of channels connectivity in space,the greater the correlation coefficient in time.