The Extraction And Recognition Of Spike And Sharp Waves In Epileptic EEG

Epilepsy is a common brain disease, which disserves human health severely, it is caused by brain damage. The discharge waves included in epileptic signal are spike, spike-and-slow wave, sharp wave, sharp-and-slow wave etc. There is paradoxical discharge in the cerebral cortex when epilepsy occurs. So the electroencephalograph (EEG) examination has a great significance for the diagnosis of epilepsy. The traditional clinical EEG examination mostly depends on expert reading the multi-channel EEG and identifying the epileptic characteristic waves to diagnose epilepsy. This method is both time-consuming and inefficient, and there is no unified standard and restrict. So the auto detection of epileptic waves is very significant.The main object of this paper is to propose some solutions with good performances based on the existing methods of signal processing. So here the epileptic EEG is analyzed from four aspects:the first step is to move the noise from the EEG, which is indispensable for the following work. What we are going to do next after the EEG’s preprocessing is identifying the epileptic EEG. Here are three steps:first, predicting the epileptic EEG by using the entropy method. But this step can only find out the approximately moment when the epilepsy occurs, it can’t determine accurately whether there is epileptic waveform. So the next step is to find out these waveforms whose instant energy is high by using empirical mode decomposition (EMD) and nonlinear energy operator (NEO) methods. And the last step is to extract the spike model from the previous waveforms and judge whether the model is a spike or not, here the popular artificial neural network (ANN) method is introduced. The experimental results show that the detecting methods here are very useful. The major contents of this thesis are as follows:Firstly, two denoising methods are studied, including independent component analysis (ICA) and wavelet transformation (WT).Then we make use of these two methods to move artifacts and burrs from the EEG, what’s more, the main information in the signal is retained. This step lays a good foundation for the next work.Secondly, the approximate entropy and sample entropy methods are discussed deeply to predict epilepsy. The data processing results show that these two complexity analysis methods have incomparable advantages and can predict epilepsy well, and the latter is better. In order to reduce the calculated quantity and remove noise, the EMD method is introduced here to predict the epilepsy combined with entropy method, then the EMD’s improved method EEMD is also induced, and the rate of mean change of the sample entropy is increased by10%, the calculated quantity is reduced further, besides, the predict result is also improved.Finally, based on the prediction of the epilepsy realized in the last chapter, the wavelet transform and NEO method are studied and used here to extracted the characteristic wave of the epileptic EEG. The waveforms whose transient energy are high are separated, and then the ANN method is used here to judge whether the waveform is a spike wave, and the mean accuracy rate is more than98%, so the recognition results are good.