Prediction Of Epileptic Seizures Based On Convolution Neural Network

Epilepsy is a chronic disorder of the brain.Abnormal discharge of the brain is the main cause of epileptic seizures.The seizure has repeatedly,characteristics of sudden attack,often manifested as loss of consciousness,the whole body involuntary convulsions,mental disorders,and patients brought great pain and pressure.If we can predict the onset of epileptic seizures before seizures,and take some necessary measures,we can reduce the burden of patients and medical staff,as well as the pathogenesis,diagnosis and treatment of epilepsy.At present,the prediction algorithm for epilepsy has made great progress,especially the application of nonlinear dynamics,but there are also many problems.As an important tool to study the characteristics of epileptic seizures,EEG reflects the seizure information that many other physiological methods can't provide.Through the analysis of epileptic EEG data,it can be found that the characteristics of epileptic EEG are distinctly different in different periods.Convolution neural network is an algorithm widely applied in various fields in recent more than 10 years,because it has good characteristics of automatic learning,especially in the field of recognition,prediction and estimation,and has made great achievements.Based on the above situation,we have studied the algorithm of epileptic seizure prediction,and proposed the automatic recognition algorithm of epileptic EEG based on convolution neural network and the epileptic seizure prediction algorithm based on convolution neural network.In this paper,we first put forward the algorithm of automatic epileptic EEG recognition based on convolution neural network.We used the convolution neural network structure to automatically learn EEG data characteristics,so as to distinguish normal EEG,epileptic seizures,epileptic seizures,EEG and epileptic interictal EEG.The experimental data of the algorithm are from the German Bonn epilepsy laboratory,which records clinical short-range EEG data.The experimental results show that the EEG automatic recognition algorithm basedon convolution neural network can achieve better recognition performance.The proposed algorithm for prediction of epileptic seizures based on convolution neural networks consists of three parts: signal preprocessing,convolution neural network and post-processing.In the pre-processing stage,we first use the mobile window technology to segment the original EEG signals,and then use the wavelet transform to get the EEG data of different frequency bands for each segmented signal.In convolution neural network,the convolution neural network structure is mainly constructed.It uses the strong feature of convolutional neural network to extract the performance automatically,and trains and tests the EEG.In the post-processing stage,the results of training and testing are further analyzed by means of logical operation and smooth processing.The experimental data of the algorithm were from the epilepsy research center of the Freiburg Medical College in Germany,and the 6leads recorded in 21 patients with long-term intracranial EEG were analyzed.The experimental results showed that 58 times of the remaining 66 seizures could be correctly predicted,and the sensitivity of the algorithm was 87.88% when the 21 epileptic seizures were used for training.Experimental results show that the algorithm based on convolutional neural network can achieve ideal sensitivity,but there is still much room for improvement in false positives.