Research On Prediction Method Of Epileptic Seizure Based On EEG Signal

Epilepsy is one of the most common brain nervous system diseases.Patients will have uncontrollable convulsions when they have seizures,and the seizure time has a strong uncertainty.It not only causes great damage to the physical and psychological of patients with epilepsy,but also hinders the normal life of patients and their families.In addition,medical means can not completely solve the troubles of all patients with epilepsy,so the prediction of epilepsy occupies an extremely important position.The current method has the defects of short prediction time and low sensitivity.Therefore,this paper studies the prediction of epilepsy by EEG,and proposes a prediction method of epilepsy:(1)Aiming at the influence of high-frequency noise and other artifacts in the original epileptic EEG signal on epilepsy prediction,firstly,the frequency of the signal is reduced,and then the band-pass filter and fast independent component analysis(Fast ICA)are used to filter the signal and eliminate the artifacts.Finally,the de-noising effect of different threshold criteria and wavelet functions is analyzed and compared,and the best choice is selected Wavelet threshold method based on unbiased risk estimation threshold criterion and adaptive threshold function is used for signal denoising.(2)Considering that the repeatability of EEG channel information will affect the performance of epilepsy prediction methods,a EEG channel selection model based on correlation analysis is designed.Pearson correlation coefficient between two channels of each patient was calculated,and then the EEG channels with repeated information were screened by channel selection criteria.The results show that EEG channel selection can accelerate the prediction of epilepsy without losing the performance of the method.(3)In order to analyze the EEG signal before seizure more comprehensively,the feature of EEG signal is extracted from time domain,frequency domain and spatial domain.Time domain features include root mean square value,absolute mean value and zero crossing points;frequency domain features are extracted by empirical wavelet transform for multi-mode decomposition and single mode selection based on Welch power spectrum,and then the instantaneous frequency and weighted frequency of signal are extracted by Hilbert transform for single mode of each channel;spatial features are extracted by improved CSP algorithm(including regularized CSP and weighted regularization CSP algorithm).The results show that the signal of interictal and pre seizure can be distinguished by single time domain,frequency domain and spatial domain features.(4)In order to effectively identify the EEG signals of epilepsy between and before seizure,the EEG signals of epileptic patients are classified by SVM classifier,and then the epilepsy warning is carried out by Kalman filter.The results show that the average prediction rate is 79.15%,and the method proposed can achieve epilepsy prediction.