The Study And Analysis Of EEG Features

A plentiful of information about physiology and pathology is contained in Electroencephalogram (EEG) signals. Some evidences of clinical diagnosis and an effective measure of adjuvant therapy of certain brain diseases are provided for doctors by processing the EEG signals. The studies about the extraction and analysis of EEG features have made some important progress and achievements at home and abroad. By using modern signal processing methods, such as sample entropy, bispectrum and 1.5-dimensional spectrum, we study the wealthy information hidden in EEG after reading a number of literatures about EEG at home and abroad. Our results can provide theoretical reference worthiness for clinical diagnose of the EEG. The main contents in this paper can be summarized as follows:(1) The basic knowledge of EEG signals is summarized, and the development, date collection, classification and other knowledge of EEG are reviewed. Moreover, some modern methods of EEG, such as time-frequency analysis, Higher-Order spectral analysis, non-linear analysis and artificial neural network analysis are introduced. Especially, some applications of wavelet analysis, bispectrum analysis, complex analysis and neural network analysis to the EEG signals processing are reviewed.(2) For shortcomings of approximate entropy algorithm, sample entropy is introduced, which is a modified algorithm based on approximate entropy, and is used to analyze EEG signals of epileptic patients and normal people. The results indicate that, on the whole, the values of sample entropy of epileptic patients are lower than those of normal people; the value of epileptic patient being in the attack period declines obviously than that of epileptic patient being in pre- attack period, and the value returns to previous level after seizure. These results are basically consistent with the symptoms of epileptic patients, which can provide reference value for the clinical diagnosis of epilepsy.(3) The kurtosis and skewness are computed to study the characteristics of non-linear and non-Gaussian of EEG signals under the different states. By using direct method of bispectrum estimate to analyze the EEG signals under three different states, there are some differences of bispectrum structures of the EEG signals under the three states. This verifies that the bispectrum analysis is an effective way of nonlinear analyses to extract the wealthy high-order information. And it contributes to the automatic classification of the EEG signals and provides the more useful information for clinical EEG studies. (4) For deficiency of traditional bispectrum analysis, a new method—1.5-dimensional spectrum analysis is adopted to analyze EEG and numerically verify the algorithm. Our results show that, the analysis of the 1.5-dimensional spectrum can so effectively inhibit the additive Gaussian noise in the signals that can easily extract useful non-Gaussian signal. Moreover, this method can reveals the quadratic phase coupling characteristic of the EEG and can greatly reduce computational capacity and complexity, and also can effectively extract useful information, which cannot be acquired by using conventional spectral analysis.