| Epilepsy is the second most common neurological disease.Seizures can cause serious damage to the nervous system of patients.EEG signal research is widely used in the medical field,and the analysis of EEG signal is the most effective detection and treatment method.This study is based on the multi-channel EEG signals of patients,focusing on the prediction and alarm of epileptic patients who are about to attack based on deep learning,so that patients can take corresponding protection measures before epileptic seizures and avoid injury.The main contents of this project are as follows:(1)A new classification method of epileptic EEG signals based on deep learning is proposed.Firstly,the epileptic EEG signal is converted into power spectral density energy diagram(PSDED),and the features of epilepsy are extracted automatically by deep convolution neural network(DCNN)and transfer learning technology.Finally,the epileptic state is divided into interictal period,preictal 30 minutes,preictal 10 minutes and seizure period.The training results show that this method has achieved good results in EEG of all subjects.The average Acc,Sen and Spe of the four categories of epilepsy are 95.025%,94.9% and 98.1%,respectively.(2)The bidirectional coupling of multi-channel epileptic EEG signals was analyzed by multi-scale symbolic permutation transfer entropy method.In order to further study the synchronization relationship of the whole cerebral cortex,the multi-channel EEG signals were divided into occipital,parietal,frontal and temporal regions.S-estimator was introduced to analyze the synchronization of multi-channel EEG signals.The results show that the results obtained by using multi-scale symbolic permutation transfer entropy are in line with the actual situation,and the analysis of information transmission between channels has a better effect;the synchronization intensity of occipital,parietal and frontal regions related to human visual,somatosensory and mental functions in seizure period is generally weakened,while that of temporal lobe related to auditory function is enhanced.(3)Based on the multi-scale symbolic permutation transfer entropy,the important EEG channels of patients with epilepsy were screened.The PSDED and the synchronous matrix diagram were fused.The deep convolution neural network was used to automatically predict the seizures of patients with epilepsy.The final classification accuracy rate could reach 96.825%.The seizure prediction horizon(SPH)and seizure occurrence period(SOP)were set.When SPH is 10 minutes and SOP is 10 minutes,the prediction sensitivity can reach 96.66% and the false detection rate can reach 0.03/h.When SPH is 30 minutes and SOP is 10 minutes,the prediction sensitivity can reach 93.17% and the false detection rate can reach 0.05/h.The analysis of the coupling and synchronization of EEG signals in epileptic patients and the method combined with deep learning provide new ideas for epileptic seizure prediction,focus localization and adjuvant treatment. |
