Research On Epilepsy EEG Signal Analysis Method Based On Neural Network

Epilepsy is defined as a nervous system disease caused by abnormal brain activity.During the period of non epileptic seizures,the performance of patients with epilepsy is no different from that of ordinary people.When epilepsy appears,the visual manifestations of epilepsy patients are limb spasm,mouth foaming,etc.At the same time,epilepsy may also lead to patients with boredom,depression,and even cause sudden death.Electroencephalogram(EEG)is an important means to detect and record brain activity.However,the analysis method of EEG in clinic mainly depends on the artificial diagnosis of doctors or experts.Therefore,there will be some problems such as low efficiency of diagnosis,and the accuracy of diagnosis needs to be improved.In order to solve the above problems,band-pass filter is used as EEG preprocessing method.Secondly,information entropy,permutation entropy,approximate entropy,fuzzy entropy and sample entropy are used to calculate the eigenvalues of the filtered EEG signals,so as to construct the feature vector for classification.Finally,the classification algorithm of epileptic EEG based on ANFIS and the prediction algorithm of epileptic seizure based on transfer learning are proposed.The main contents of this thesis are as follows:(1)Six groups of fourth-order band-pass filters with different frequency intervals were used to preprocess all the test data sets.Because in the process of EEG signal acquisition,there are often mixed with other frequency bands of signal interference,and the frequency range of different waveforms in EEG signal is different.Therefore,in order to get a better classification effect,this thesis first preprocesses the original EEG signal in the form of filtering.(2)Information entropy,fuzzy entropy,permutation entropy,approximate entropy and sample entropy are used to calculate the feature vector of EEG data.(3)ANFIS is used as a classifier to classify the constructed feature vectors.This method achieves 95.60% classification accuracy in CHB dataset and 97.73% classification accuracy in Bern dataset.(4)Vgg19 network is used as the basic classifier,and the idea of transfer learning is added.Because the traditional deep learning has serious disadvantages of data dependence.Moreover,in the actual clinical treatment,there are some problems such as high cost of EEG signal acquisition and difficulty of labeling.The idea of transfer learning is to train a network model in a field with enough data and transfer the knowledge or network structure to another field.It is very suitable for the training of small sample data.The algorithm takes the Epilepsy EEG data set of Boston Children’s Hospital as the experimental data,six groups of band-pass filters with different frequency intervals as the preprocessing method,and the sample entropy as the feature selection.Finally,VGG19 network after transfer learning is used as classifier to recognize epileptic EEG signals.The final experimental results show that the longest average prediction time of this algorithm is 41.30 minutes,and the average prediction time of all patients is 23.82 minutes;the highest prediction accuracy rate is 93%,and the average prediction accuracy rate is 86.4%.The lowest false positive rate was 22%and the average was 34%.The final experimental results show that the method is reliable for epileptic seizure prediction.