Using Dictionary Pair Learning For Seizure Detection

Epilepsy is a chronic mental illness of abnormal nerve function caused by paroxysmal abnormal discharge of brain neuron group,which is transients sudden and repetitive.Once an epileptic seizure occurs,the patient will lose consciousness,with body convulsions and mental disorders,etc.,and may even leads to death or disability.Therefore,epilepsy will not only affect the normal work and life of the patient,but also bring a burden to the society.According to statistics,about 80%of the epileptic patients have shown an abnormal electroencephalogram(EEG).The main manifestations were spinous wave,sharp wave,spinous-slow wave,sharp-slow wave,multi-spinous wave,etc..So far,EEG is an important auxiliary tool for the study of epileptic seizures and the diagnosis and treatment of epilepsy.The diagnosis of epilepsy is mainly done by experienced medical workers through visual analysis of the electroencephalogram.The analysis of a large number of EEG by eyes is a complicated and time-consuming work with low efficiency and certain subjectivity.The automatic detection of epileptic EEG by computer is a valuable diagnostic method.It can not only reduce the burden of medical personnel,improve the diagnostic efficiency of epileptic seizures,but also help deepen the understanding of the internal mechanism and rule of epileptic seizures.The traditional method of seizure detection is to carry out the feature extraction first and then classified by a classifier,and the feature extraction depends mainly on human selection,which is a difficult task.The development of portable medical equipment has also brought challenges to the research of epileptic detection algorithms.Therefore,we need to propose an automatic epileptic detection algorithm to meet the real-time online requirements,transform the research results into practice to effectively control and treat epileptic diseases.Therefore,this paper proposes a new method of detecting seizure for long-term intracranial EEG based dictionary pair learning algorithm,which does not require complicated feature extraction algorithm.It can greatly reduce the detection time and is conducive to the real-time implementation.In this paper,the proposed algorithm using the dictionary pair learning for seizure detection is mainly divided into preprocessing stage,training stage and testing stage of the dictionary pair learning model and post-processing stage.First of all,the original EEG data are segmented,and then the EEG epochs are preprocessed by wavelet filtering,differential operation and kernel function,by which the interference noise of the original EEG data can be filtered out and the difference between the seizure EEG and the interictal EEG becomes more significant.This stage is to make preparations for the classifications.Next,we randomly select seizure EEG and interictal EEG as training samples,and then use the trained dictionary pair to make sparse coding for testing samples and calculate the reconstruction errors.In the dictionary pair learning model,synthetic dictionary and analytic dictionary is obtained from the training samples by alternating minimization method,and sparse coefficients can be obtained by using linear projection instead of high cost l0-norm or l1-norm optimization,thereby greatly reducing the algorithm running time.Finally,the difference between the sub dictionary pair of seizure EEG and interictal EEG is taken as the decision value,and post-processing is performed to improve the recognition rate and reduce the false detection rate of the system.The evaluation experiment of the algorithm used 530 hours of data of 20 epileptic patients with 81 seizures.The detection method proposed in this paper achieved a segment-based sensitivity of 93.39%,a specificity of 98.51%,and an event-based sensitivity of 96.36.%,false detection rate was 0.236/h.