Automated Classification And Detection Of Epilepsy EEG Based On Multi-feature Extraction

Epilepsy is a typical transient neurological disorder of brain function.When epilepsy occurs,EEG signals produce a large number of slow-spinning characteristic waves.Clinically,the epilepsy disease is diagnosed mainly by visual analysis of the patient’s EEG,but this method is time-consuming,inefficient,subjective,and easy to cause misdiagnosis.Therefore,it is of great significance to study the automatic detection of epileptic EEG signals.In this paper,an automatic classification detection method for epilepsy EEG based on multi-feature extraction is proposed.The feature extraction algorithm is used to extract the signal characteristics of five basic rhythms of EEG,and the classification algorithm is used to realize the automatic classification detection of epilepsy EEG.Based on the work of predecessors,this paper has mainly completed the following work:1.The research status and future development trends of EEG signal feature extraction and automatic classification algorithms are analyzed.The characteristics of epileptic EEG signals are briefly introduced.2.Introduce two correction operators for the wavelet aliasing frequency aliasing phenomenon,and eliminate other frequency components attached to the wavelet packet subband.The experimental results show that the introduced operator can better eliminate the frequency aliasing of the reconstructed signal,so that the five basic rhythms of EEG can be accurately extracted.3.An algorithm for extracting EEG based on the fast and slow wave energy ratio and sample entropy of epileptic EEG rhythm waves is proposed.9 fast and slow wave energy ratios and 5 rhythm wave sample entropies of epilepsy EEG were extracted as feature quantities for classification.According to the analysis of variance analysis,the p values of the 14 features extracted from the epileptic brains were less than 0.0001,and there were significant differences,which could effectively reflect the clinical pathological information of epileptic EEG4.Using the "one-to-one" voting method to improve the multi-classification application of the SVM classifier,combined with the grid optimization and cross-validation method to select the best disciplinary factor C and nuclear radius σ of the SVM classifier,and verify by the Fisher_iris data set.Improve the feasibility of the algorithm.The experimental results show that the improved method achieves 100%classification accuracy when C and σ take 2 and 0.7 respectively,and successfully achieves multi-class prediction.5.The feature extraction algorithm and classifier algorithm proposed in this paper are used to solve the classification problems of 9 common epilepsy EEG signals in 4 categories,and compared with probabilistic neural network algorithm and K-nearest neighbor classification algorithm,the average classification accuracy rate is above 98.67%.The results show that the proposed method based on multi-feature extraction for epilepsy EEG automatic classification detection has better classification and recognition of the above four types of epilepsy EEG signals.