Prediction And Detection Of Epilepsy Based On Neural Network

Epilepsy,as a neurological disease,loses consciousness and convulsions during a seizure,and the patient may suffer other injuries.Therefore,an effective algorithm for predicting epilepsy is of great significance for protecting patients.Epilepsy EEG data is generally recorded and marked by experienced doctors.This method is not efficient and relies heavily on work experience.An automatic epilepsy detection algorithm is of great significance for improving the collection efficiency and quality of epilepsy EEG.First,the time-frequency features extracted in the existing prediction algorithms are not obvious,and the prediction results are inaccurate.In order to improve this problem,this paper studies an algorithm for epileptic seizure prediction based on short-time Fourier transform and continuous wavelet transform time-frequency features.The prediction algorithm fuses the two time-frequency feature matrices into a new feature matrix,and uses a convolutional neural network to classify the feature tensor.Finally,the statistical classification results are used to predict whether the patient has seizures.The average sensitivity of the algorithm is effectively improved,and the average error prediction rate is reduced.Secondly,improve the convolutional neural network structure and training algorithm in the prediction algorithm.In view of the small amount of pre-seizure data and low requirements for the classification network structure,the algorithm selects three convolutional layers to extract image features.In order to further improve the classification accuracy,the convolutional layer selects Rectified Linear Unit(Re LU)function as the activation function to further improve the classification accuracy.In the network training phase,an adaptive moment estimation(Adam)optimization algorithm is selected to train the network.Finally,to address the problem of low detection efficiency caused by excessive reliance on manual detection,this paper studies an automatic epilepsy detection algorithm based on Bidirectional Gated Recurrent Unit(Bi-GRU)and Local Mean Decomposition(LMD).The algorithm uses the local mean decomposition method to decompose the original EEG signal,extracts the statistical features of the decomposed components,uses the bidirectional gated recurrent network to classify the feature data,and finally detects epilepsy based on the classification results.The algorithm effectively improves the average sensitivity and average specificity.