Time Frequency Analysis And Its Application In EEG Signal

Epilepsy is often the sign of neurological disease or dysfunction, it disserves human health severely. The waves of epileptic discharge include spike, sharp wave, spike-and-slow wave, sharp-and-slow wave etc. And the important auxiliary diagnosis of epilepsy is based on electroencephalograph (EEG). During the traditional clinical EEG examination, it mostly depends on expert reading the multi-channel EEG and detecting the epileptic characteristic waves to diagnose epilepsy. Because this method easily brings some subjective selectivity, moreover, in some cases the data are very large and this task looks very onerous and vapid, so the auto detection of epileptic waves is very significant.The traditional approaches for analysing EEG signals are in time domain or in frequency domain. The time domain methods are very intuitionistic, and in frequency domain the rhythm and spectra characteristics are considered. But both methods are all based on the assumption of stationarity of EEG signals. Therefore, the detection result is not quite reliable due to EEG signals are significantly nonstationary and multi-component. But time-frequency methods fit to analyze nonstationary signal, it can represent the distribution of signal's energy at each time in frequency. So time-frequency analysis has better future in the EEG signals processing.This thesis applies time-frequency analysis to epileptic EEG signal, and extracts the useful time-frequency characteristics, so as to complete the automatic detection of epileptic characteristic waves and alleviate the heavy labour force of the doctors and improve the efficiency of diagnosis. Firstly, this paper studies the basic theory and overviews the existing time frequency analysis methods and summarizes the development of suppressing the cross-terms in recent years. A new time frequency distribution based on auto-regressive model (AR) spectrum is given and the order of AR model is made optimum. Secondly, this paper gives two analysis methods based on empirical mode decomposition. One is a time-frequency distribution based EMD, the other is nonlinear energy operator (NEO) based on EMD, and both of them have good results in epileptic EEG signal processing. Thirdly, two synthetic methods of spike detection are proposed. One is based on time frequency analysis of Choi-Williams distribution (CWD). In this method, it employs the way of singular value decomposition (SVD) to get a better time frequency plane, then applies time frequency divergence measures to complete automatic detection. The other method combines time-frequency analysis and Jensen function to complete spike detection, and extracts thecharacters of spike wave and applies artificial neural network (ANN) to do further detection, so as to reduce the false detection ratio. Finally, this thesis makes a system of epileptic characteristic waves by matlab computer language, several detection methods are aggregated in this system, including the methods of this thesis, and also the ones of other literatures.