| Electroencephalogram(EEG)can record the abnormal discharge of brain neurons in patients with epilepsy.Clinicians can visually recognize these abnormal discharges based on experience by reading the patient’s EEG,and use them as a reference for diagnosis.However,in order to record effective abnormal discharges,patients often need dozens of hours of EEG examinations,and doctors have to spend a lot of time reading these long and boring EEG,resulting in inefficient diagnosis procedures.Therefore,it is of great significance to use an epilepsy automatic diagnosis system to reduce the burden on doctors and improve diagnosis efficiency.This paper studies the characteristics of EEG in epileptic patients and designs an intelligent diagnosis system based on scalp EEG,which includes three modules:(1)EEG pattern recognition module: Based on the study of the characteristics of EEG data of patients with epilepsy,this paper proposes an automatic EEG recognition algorithm based on convolutional neural network.The algorithm first uses wavelet denoising to preprocess the EEG data,which can effectively remove some EEG artifacts without changing the epileptic EEG characteristics.Then the one-dimensional convolutional neural network model built in this paper is trained and tested on the preprocessed EEG data,and the hyperparameters of the convolutional network are optimized using Bayesian optimization algorithm.Experimental results show that on a data set containing four types of EEG segments,the EEG pattern recognition algorithm based on convolutional neural network proposed in this paper achieves an accuracy rate of 96.67%,which can effectively identify epileptic-specific discharge pattern.(2)EEG synchronization analysis module: The synchronicity of epilepsy-specific discharges has a reference value for lesion location.The correlation analysis method can be used as an evaluation index of the synchronization of time series,but the existing correlation analysis method cannot combine the characteristics of epilepsy-specific discharge to analyze its synchronization.This paper proposes a local correlation coefficient based on salient attention as an indicator of epilepsy-specific discharge synchronization.Experimental results show that this indicator can significantly distinguish epilepsy-specific synchronous discharge and asynchronous discharge.(3)Active learning module: According to the characteristics of epilepsy EEG data,this paper designs an uncertainty sampling strategy based on principal component analysis and density peak clustering.And use a data set containing 6000 samples to compare with random sampling strategy.The experimental results show that,In the case of using the same amount of training data,the recognition accuracy rate of proposed active learning strategy on the reserved test set and the remaining unlabeled sample set is 2.09% and 10.04% higher than the random sampling strategy.In the case of achieving the same accuracy rate,the training samples used are reduced by 37.5%and 57.5%.It proves that the sampling strategy proposed in this paper can select samples with more learning value for the training of epilepsy EEG recognition model,improve the recognition effect of the model,and significantly reduce the cost of sample labeling. |
