Seizure Prediction Algorithm Based On Spike Rate And AR Modle In EEG

The epilepsy is a kind of the dysfunction of the brain, which is mainly due to the super-synchrony discharging of the cerebral cortical neurons. More than50million people in the world are diagnosed with epilepsy, and its recurrent attacks seriously hurt patients’physical and mental health. Epilepsy prediction will not only help patients to take preventive and protective measures, but also contribute to the in-depth exploration of the pathogenesis of epilepsy and epilepsy diagnosis and treatment technology. Thus, many scholars dedicated to the study of epilepsy prediction at home and abroad.Spike wave is a sudden change of EEG with obvious difference from background. It is one of the most characteristic performances of epileptiform discharges. The emergence of spikes indicates that there are irritating lesions in the brain. So in clinical epilepsy checks, the most important thing is to observe whether there are spikes in EEG. Autoregressive model is a linear prediction method which can be applied to analyze EEG. Assuming the EEG signal can be described by a linear filter and can be approximated by the AR process, select the appropriate order and parameters of the AR process can approximate EEG as much as possible. Therefore, this paper analyzed the rule of spike rate changes and AR parameters changes in different EEG period, and then proposed a novel seizure prediction method based on spike rate and AR parameters.For the most of linear classification algorithms, they can get the test sample classification label, but cannot get the probability of the test sample belonging to a certain category. Bayesian Linear Discriminant Analysis (BLDA) is based on the evidence framework of Bayesian regression. It is widely used because that it can get the test sample classification label and the probability of the test sample belonging to a certain category. BLDA can be regarded as the expansion of Fisher Linear Discriminant Analysis (FLDA). Compared with FLDA method, BLDA uses regularization method to solve the high-dimensional data accompanied by the noise of over-fitting problem. This article will use BLDA as a linear classifier.Preprocessing is first done on the EEG signal to filter out high-frequency interference and work frequency interference. Then EEG signal is segmented with a moving window. The spike rate and AR parameters is calculated on each EEG segment as feature vectors. Then BLDA is used to do classification and discrimination and the final classification results are filtered by Kalman filter to smooth the classification results.The test data used in this article comes from the Epilepsy Research Center of Germany Freiburg Medical School, including a total of87seizures of21epileptic patients’intracranial EEG data. Experimental results showed that, removal of40seizures for training, there are42seizures in the remaining47seizures are correctly predicted, and the sensitivity reached89.36%with a false prediction rate of0.09/h.By comparing our proposed algorithm with increments of accumulated energy and the phase synchronization based on the wavelet transform, we can conclude that the proposed algorithm is an effective epilepsy prediction algorithm.