Epilepsy Detection Based On Bayesian Convolutional Neural Network

Epilepsy(Epilepsy)is a common neurological disease,and its main cause is the abnormal discharge of brain neurons.Its main feature is recurrent seizures,which are long or short severe withdrawal symptoms.The definition of epilepsy is the patient’s continuous repetitive seizures without predisposing cause.It usually causes short-term brain dysfunction,has become the second most common disease in neurology after headache,and seriously endangers the physical and mental health of patientsAs an effective method for clinical diagnosis of epilepsy,electroencephalogram(EEG)plays an irreplaceable role.When the brain is active,a large number of neurons generate potentials synchronously.It can accurately record the spontaneous bioelectric signal from various parts of the scalp through more sophisticated electronic equipment,and then amplify it through an amplifier to record the obtained waveform.However,the detection of epilepsy through EEG often requires medical staff to judge with the naked eye.Usually the clinical long-term EEG needs to be performed continuously for about 24 hours.The judgment of such a long EEG not only requires a lot of time and energy,but also the doctor’s ability and energy in the judgment are very different,and it is also very strong.Subjectivity.Especially in some atypical EEG waveforms,there may be big differences.With the popularization of information technology and the research of various machine learning algorithms,the automatic detection of epilepsy has also made great progress.From the initial half-wave EEG feature extraction,it gradually evolved to the later time domain,frequency domain,time-frequency domain,and non-frequency domain.The further improvement of linear feature extraction methods and the combination of support vector machine(SVM),random forest(rf)and other machine learning algorithms to classify epileptic signals greatly improves the accuracy of classification.This is currently a more common one.Kind of detection method.In recent years,with the advancement of artificial intelligence technology,various deep neural networks such as artificial neural networks(NN),long and short-term memory networks(LSTM),convolutional neural networks(CNN),etc.have been applied to the field of epilepsy detection.This paper proposes an epilepsy classification and detection algorithm based on s-transform and Bayesian convolutional neural network.S transform is a feature extraction method in the time-frequency domain.It has a direct relationship with Fourier transform,and its resolution is also different at different frequencies,which is closely related to wavelet transform.The Bayesian convolutional network is a variant based on the convolutional neural network,which introduces the probability model into the weight of the deep neural network,adds uncertain factors to the field of epilepsy diagnosis,and makes the classification network more in line with the cognition of the human brain Happening.This method first preprocesses the signal through s transform,and obtains the time-frequency domain characteristic map of the signal.Then use Bayesian convolutional neural network for classification,and finally use smoothing filter and collar technology for post-processing.This method obtained an average sensitivity of 99.03%,a specificity of 97.74%and an accuracy of 97.75%in a total of 509 hours of EEG data from 21 patients.The sensitivity and specificity are as high as 100%and 99.94%,respectively.The sensitivity of event detection is 96.67%,and the error detection rate is 0.38.It provides new methods and ideas for epilepsy detection based on artificial intelligence.