Classification Of Epilepsy Signals Based On Wavelet Transform And Sample Entropy

Epilepsy has been a serious problem to the medical profession for a long time.The complexity and suddenness of epileptic EEG signals bring great challenges to the detection of epilepsy.In general,epilepsy detection includes three processes:signal preprocessing,feature extraction,and classification.For the automatic detection system for epilepsy,the accuracy of identifying epilepsy and the accuracy of predicting epilepsy are of great significance for the treatment and the rehabilitation of patients.In this paper which based on the feature extraction and classification identification of epileptic EEG signals,the following studies are mainly carried out:(1)A feature extraction method based on discrete wavelet for epileptic EEG signals is proposed.First,the wavelet transform is introduced for the shortcomings of the Fourier transform.Then the advantages and disadvantages of continuous wavelet transform and discrete wavelet transform for EEG signals are compared.The continuous wavelet transform has high redundancy in decomposing epileptic EEG signals,discrete wavelets are used in this paper for experiments.Finally,five layers of discrete wavelets are used to decompose epileptic EEG signals to extract their time-frequency characteristics.(2)An EEG signal feature extraction method based on discrete wavelet-fast sample entropy is proposed.The eigenvalues that best reflect the characteristics of epileptic EEG signals and the complexity of EEG sequences are extracted by the combination algorithm.An optimized sample entropy,named fast sample entropy algorithm,is proposed.Then the sample entropy is calculated for the EEG original signal to verify the efficiency of the two sample entropy algorithms.The results show that the effect of the fast sample entropy algorithm is significantly better than the original sample entropy algorithm from the perspective of time cost.Finally,the fast sample entropy of the detailed signals of each group from EEG data is calculated.The experimental results show that the sample entropy distribution of the details of the brain waves from healthy volunteers is quite different from the sample entropy distribution of the details of the brain waves from patients with epilepsy.It is found that EEG signals from healthy volunteers and patients can be classified by the sample entropy algorithm.(3)The classification of epileptic EEG signals based on support vector machine and naive Bayes algorithm is proposed.Firstly,the evaluation indexes of the machine learning algorithm for classification experiments are introduced.Finally,the classification results are obtained by inputting different features:time-frequency characteristics,nonlinear dynamics and mixed features into two machine learning classifiers.The classification results of different features and the capacity of two different classifiers are compared.The experimental results show that the hybrid features of time-frequency features and nonlinear dynamics can better reflect the characteristics of epileptic EEG signals,and the classification effect is better than single time-frequency feature or nonlinear dynamics feature.