Nonlinear Analysis For Epileptic Brain Electric Signals

Epilepsy disease seriously threatens the health of patients. Analysis of brainelectrical signal is an important approach for epilepsy study. At present, intracranialelectrocorticogram (ECoG) of single electrode or double electrodes in focal area is themain research data for epilepsy analysis. There are two pieces of data which is studiedin this thesis. One part of data was taken from the designed electroencephalogram(EEG) acquisition experiment. This part of data includes20electrodes of scalp EEGcollected clinical. The other part is the ECoG data provided on line. In this thesis,nonlinear analysis methods of complexity, synchronism and multi-scale are applied toEEG and ECoG. The effective features that can not be extracted by linear methods areobtained.The complexity analysis of LZ complexity and correlation dimension of phasespace was used to analyze the epilepsy brain electrical data. The complexity feature ofbrain electrical signals was extracted. The results show that the complexity ofepilepsy EEG is lower than that of healthy EEG, and that complexity of ECoG duringa seizure (ictal ECoG) is lower than that of a seizure-free interval (interictal ECoG).The complexity can be used as features to auto-diagnose and predict epilepsy.The method of order recurrence plot was used to analyze the determination ofepileptic brain electrical signals. The results show that the determination feature isconsistent with complexity feature analysis. Nevertheless, the computation speed ofdetermination is much faster. It is more appropriate for the brain electrical signals ofshort time, high noises, and high nonstationarity. Base on the point that brain is anintercoupling and interacting complex net work, permutation mutual information waspresented to study the information transmission of different electrodes.Multi-electrodes synchronism analysis is used to EEG data. The results show that theexchange of information of different brain regions of epilepsy patients is obviouslymore than that of healthy objects. That is the synchronism of epilepsy patients isobviously enhanced.Different rhythms of brain electrical signals are obtained by waveletdecomposition and reconstruction techniqes. Feathures of wavelet entropy,complexity and determination were analyzed in different rhythms. The results showthe changing rules of these feathures in different rhythms are not identical. By independently analyzing subbands, more accurate information underlying the epilepsybrain electric signals can be extracted.Neural network based on unscented Kalman filter(UKF) was used as an classifierof the features of epilepsy brain electric signals. The algorithm of UKF can estimatethe weights of neural network at a high rate of speed. So the problem the lower speedof net training is solved. The results show the classification performance of theclassifier is superior to linear discriminant analysis(LDA).In this thesis, multiple features were extracted from epilepsy brain electric datataking advantage of nonlinear analysis methods. A theoretical basis is provided forauto-diagnosis of epilepsy and seizure prediction.