A Study On Seizure Prediction Based On Single Channel And Brain Network Analysis Of EEG Signals

Epilepsy is a common chronic brain disease,affecting more than 50 million people worldwide.Seizures often occur suddenly and have severe consequences,greatly affecting the patients’ normal activities and even endangering their lives.With the development of artificial intelligence,it has become possible to combine EEG signals with computer technology to achieve automatic seizure prediction.Automatic seizure prediction allows the patient or medical caregiver to take measures in advance to prevent the damages caused by seizures.Feature extraction is a key step in seizure prediction,but most of the features used in seizure prediction studies do not take into account the characteristics of the epileptic EEG signal and brain diseases.Considering this deficiency,this thesis extracts features matching these characteristics from single-channel and brain network dimensions based on EEG signals of epilepsy patients,to analyze the differences in patients’ brain patterns during different periods of epileptic seizures,and combines them with machine learning technology for seizure prediction.The main works of the thesis are as follows:First,based on the ability of nonlinear feature analysis to explore the EEG signals of epileptic patients,and the unique advantages of fuzzy entropy,this thesis proposes the use of fuzzy entropy,a nonlinear indicator,as a single-channel EEG feature to achieve seizure prediction.Moreover,considering the problems that exist with the use of all channels for EEG monitoring and seizure prediction,the relevant channels are selected in this paper as the optimization method for singlechannel analysis.That is,considering the onset principle of epilepsy patients,the intensity of phase amplitude coupling enhancement is proposed as a feature to select the relevant channels.The results show that fuzzy entropy can well separate the EEG signals in the pre-and interictal phases,and this feature combined with the SVM classifier can achieve an average prediction time of 27.3 minutes and an average prediction accuracy of 96.75%.After implementing the channel selection scheme,an average of 6.4 channels were selected for each patient and an average prediction accuracy of94.84% is obtained.These experimental results suggest that this protocol is promising for application in the development of portable seizure warning devices.Then,since epilepsy is an abnormal brain network disease,in order to further investigate the relationship between channel signals representing different brain regions of epilepsy patients and analyze the differences of brain networks under different states,this thesis proposes to use transfer entropy,a nonlinear directed connection indicator,to calculate the causal connection relationships between channel signals to establish the effective brain network for further analysis.After obtaining the transfer entropy matrix of the patient,the threshold processing is applied to obtain the causal brain network.Then,the differences in brain networks under different states are analyzed by combining graph theory and statistical methods,and a feature combination scheme is designed.Finally,the feature combination is fed into the SVM to achieve seizure prediction.The experimental results show that this approach can not only achieve an average accuracy of 94.25% for all patients in their optimal frequency bands using relevant network features,but also effectively explore the differences in patients’ brain networks during epileptic seizures,which may help to understand the mechanisms of seizure onset.In conclusion,the seizure prediction schemes proposed in this thesis are designed from two different dimensions,which provide new insights for seizure prediction and relevant treatment.The proposed schemes can not only be used to analyze changes in brain activity of epilepsy patients during seizures,but also achieve excellent seizure prediction performance.This work is expected to be applied to develop portable seizure warning systems for epilepsy patients,and to help biologists explore the mechanisms of changes during epileptic seizures.