Denoising Sparse Autoencoder Based Classification Of Epilelptic Seizures In EEG

Epilepsy is an abnormal function disease of movement,consciousness and nerve caused by abnormal discharge of brain neurons in the brain.In the case of epileptic seizures,the patient has a sudden physical convulsion and loss of consciousness,which brings great physical and psychological pain to the patient.Therefore,the prevention and treatment of epilepsy is of great significance for people with epilepsy.Electroencephalogram(EEG)plays an irreplaceable role in the diagnosis and treatment of epilepsy.The doctor judges the patient’s condition by observing the EEG of the patient.However,the workload of the doctors’ observing the EEG is very large,consuming time and physical strength,and it is very easy to misjudge.With the development of computer technology,doctors can mark the epileptic EEG by means of automatic seizure detection,and then make a diagnosis.Automatic seizure detection can not only help doctors improve the accuracy of diagnosis and treatment,but also save time greatly,so the research of automatic seizure detection is of great value for the prevention,diagnosis and treatment of epilepsy.In this experiment,we propose a new automatic epilepsy classification system.This system is a deep neural network based on the denoising sparse autoencoder(DSAE).The denoising sparse autoencoder is based on the autoencoder(AE)network.The sparsity constraint and corruption operation are added to the hidden layer and input data of autoencoder network respectively,and then the denoising sparse autoencoder is obtained.The sparsity constraints in the network make most of the neurons in the hidden layer be in a suppressed state,which simulates the transmission of neurons in the human brain.Sparsity constraints can obtain higher-level and more effective expression of input data,which helps to obtain more accurate classification.As a result,by adding a corruption operation to the input signal,the network must learn to remove these noises to obtain an input that is free from noise pollution.This allows the system to learn the input signal to be more robust and improve the generalization ability of the system.The flow of the epilepsy EEG classification system proposed in this paper is as follows:First,the original EEG signals are preprocessed,and the preprocessing includes data segmentation and normalization.The data segmentation can increase the number of data segments to make the EEG signal easier be analyzed and calculated.The normalization method uses the Z-score standardization method to make the amplitude of the EEG signal match the detection network.Then,the preprocessed EEG signals are input as training data into the denoising sparse autoencoder network.The network automatically adjusts the parameters through the learning algorithm so as to learn the characteristics of the training data to obtain a good expression of the input data.We set a logistic regression classifier at the top-level coding layer to associate the effective expression of the signal obtained by the network with its category,that is,classify input samples.Finally,the classification results of the classifier are postprocessed,and the final classification results are obtained.We used a five-sets epilepsy EEG database to test the classification system proposed in this paper and obtained a satisfactory classification result.The average sensitivity,specificity and recognition accuracy of two class problem and three class problem were all 100%.The average accuracy in the five class problem was 92%.The experimental result shows that the EEG classification system proposed in this paper can effectively classify epilepsy EEG,which has certain feasibility and clinical application value.