Bi-LSTM Seizure Prediction Analysis Based On Functional Brain Network

Epilepsy is a high incidence of stroke functional syndrome caused by excessive release of brain neurons.Its sudden and repetitive characteristics place a great burden on patients both physically and psychologically,leading to depression,impairment of consciousness,and suicidal behavior.According to a survey report by the International Health Organization,the number of epileptic patients in the world is currently about 65 million,while the number of epileptic patients in China is between 9 million.About 70%of epilepsy patients in China can control their onset by using anti-epileptic drugs on time.Timely prediction of the occurrence of epilepsy and timely warning of patients can win valuable medication opportunities for patients,and even control the occurrence of upcoming epilepsy,which has significant application value.This article will describe the research work carried out from the following three aspects:(1)In response to the problem of high noise and pollution in epileptic EEG signals,in order to obtain high-quality epileptic EEG signals and improve the efficiency of epilepsy classification and prediction,this study conducted filtering,denoising,and removal of EEG artifacts on the EEG signals.Firstly,a Common Spatial Pattern(CSP)spatial filter is used to filter epileptic EEG data,with the aim of making the differences between EEG data from different periods more pronounced after filtering.Secondly,wavelet transform(WT)was used for denoising,achieving temporal segmentation of high-frequency and low-frequency EEG signals.Finally,independent component analysis(ICA)is used to remove artifacts from the EEG signal,making it smoother while retaining its original features.(2)In order to solve the problems of low separability and poor classification performance in the feature extraction process,this article establishes a functional brain network based on phase lock values to extract node degree features and cluster coefficient features,and constructs corresponding feature vectors.From the perspective of visualization and separability,the two selected features are compared and analyzed,and the separability of different features is compared,laying a foundation for future feature classification.(3)Using the Bi LSTM network model for deep learning classification of epileptic EEG features,the storage unit,number of neurons,and learning algorithm of the single-layer dual LSTM structure were optimized.The impact of different optimization algorithms on classification success rate and the impact of changing the number of neurons and initial learning rate on performance were compared and analyzed.This research not only focuses on the success rate,but also emphasizes the importance of determining hyperparameter(selection of optimization algorithm,initial learning rate,number of hidden neurons)in the Bi LSTM model.The results indicate that the optimized model not only improves accuracy but also efficiency,providing a new perspective for epilepsy prediction.