Chaos Dynamics Modeling Of EEG Signal Based On Multiple Wavelet Neural Network And Its Application

Electroencephalography, for short EEG; is an external representation of bioelectricity comesfrom the brain, and also one of the important assistant tools for clinical diagnosis. But because ofthe limitation of classical EEG analysis, no major achievement or success has been got, and theattempts to use it as a clinical diagnostic tool has not yet successful. Recently, many evidenceshave showed that electroencephalogram (EEG) signal is chaotic signal produced by the nonlineardynamics brain system rather than purely random, so a novel research method based on chaosdynamics is introduced in this paper which can reflect the essential nonlinear characteristic ofEEG signal and reveals the useful information involved in EEG effectively. The information gotfrom the chaos dynamics method can be used as the assistant material when psychiatrists makingtheir diagnosis decision. For this reason, the problem about EEG signal modeling based on thechaos dynamic is discussed detailedly in this dissertation.In general, there are two very important factors when nonlinear modeling the EEG signal:phase-space reconstruction and nonlinear mapping function approximation. Firstly, we introducesome basis conception about chaos dynamic and chaos dynamics modeling, then Lorenz chaostime sequence and real EEG signal are used to discuss the phase-space reconstruction problem,including the choice of time delayτand the minimum embedding dimensionm ; After that,which kind of neural network should be choose to approximate the nonlinear mapping functionand the optimization algorithm of neural network is discussed. Wavelet neural network is chosenfor this job for its excellent nonlinear function approximation ability and fast convergence speed,and genetic algorithm is applied to search for the optimal model parameters.We find that EEG signal is always a piecewise chaos signal comes from the piecewise chaosdynamic brain system rather than a smooth chaos signal. So a novel model based on multiplewavelet neural network is proposed to reconstruct the essential piecewise chaos dynamics ofEEG, and hidden Markov model is applied to determine the state probability at all time steps.The proposed multiple wavelet neural network model is used to predict the parameter-varyingLorenz chaos time sequence and real EEG signal respectively, the simulation results indicates that the multiple model can exploit the inherent chaos dynamics in EEG more effectively. At last,the new multiple model is used as a predictor to detect the instantaneous occurrence of epilepsyembedded in EEG by observing the abrupt change in prediction error of the mixed signalcomposed of normal EEG signal and epilepsy signal.Finally, summarized the work of this paper, and pointed out some problems about themultiple wavelet neural network model to be solved and the direction for the furtherdevelopment.