Chaos Modeling Using HMM-Normalized Radial Basis Function Hybrid Model Approach And Its Application

Electroencephalography (EEG) is an external representation of human brain's behaviors, and also an important assistant tool for clinical diagnosis. The classical EEG analysis methods are from the angle of frequency spectrum and statistics, which can make some detailed explanations of EEG and contribute to the effective application of EEG. However, these analysis methods can not have a deep and effective understanding of the nonlinear dynamics characteristics of human brain.There exist more evidences that EEG signal is typical chaos signal produced by the chaos dynamics brain system. In this research, the normalized radial basis function network (NRBFNN) is firstly introduced. According to the chaos characteristics of EEG, we propose a new method to model and predict the chaos signal, which is called HMM and normalized radial basis function network (NRBFNN) hybrid model. At the same time, this three-layer normalized RBF network is trained by Genetic Algorithm (GA) and Hidden Markov Model (HMM) is trained by Baum-Welch Algorithm. Compared to conventional single neural network model, the new model can approximate and reveal the essential piecewise chaos dynamics characteristics of the spatio-temporal chaos signal and chaos signal with varying parameter more effectively.The simulations with real EEG signal and epileptic signal all evaluated the effectiveness of the proposed model. The experimental results show this new hybrid network model can reveal the chaos dynamics of EEG. At last, the proposed hybrid network model is used to detect epilepsy in normal EEG signal based on the chaos dynamics characteristics difference between them, and the detection result shows that the proposed model is an effective epilepsy detector.Finally, we summarize the work of this paper, and point out some problems about the hybrid network model to be solved and the direction for further development.