The Research Of Epileptic EEG Signals Identification Based On EMD And Mixed Features

Epileptic is the second largest disease in the nervous system; only in China, the number of cases is as high as more than nine million. In the traditional medical diagnosis, because of the uneven distribution of medical in the world and doctors lack of experience, misdiagnosis often happens; automatic identification of epileptic EEG signals, which can effectively avoid misdiagnosis, makes the epileptic patients be diagnosed as soon as possible and get timely treatment.Automatic identification of epileptic EEG signals consists of two parts:feature extraction and feature classification, where feature extraction is the most important part. In the process of feature extraction, there are two points which are important and directly affect the accuracy of the identification result, one is whether the characteristics extracted are comprehensive; the other is whether the information can describe the characteristics of epileptic EEG signals.In this paper, a new feature extraction based on empirical mode decomposition (EMD) and mixed features is proposed. Firstly, decompose different EEG signals adaptively using EMD algorithm and get a series of intrinsic mode functions, which arrange from high frequency to low frequency. Namely, each intrinsic mode function represents the characteristics of EEG signals of different frequency bands. Secondly, reconstruct the time series. Select some intrinsic mode functions whose frequency band is corresponding with epileptic ictal one, to form a new time series. Then extract the linear and nonlinear characteristics of the new signals. They correspond to the linear and nonlinear properties of the brain electrical signal. And mix them as a new feature which is the input of classifier, after the classifier is trained, we can get the result of identification. The results of simulation experiments show that the feature extraction using proposed algorithm can better describe the characteristics of epileptic EEG signals, and easier to classify. ELM is selected as the classifier because the training time is short and the classification accuracy is high, and higher than other similar research.