Research On Epileptic Seizure Detection And Prediction Model

Epilepsy is a chronic brain disorder syndrome caused by multiple factors.The repetitive nature of seizures causes great physical and mental suffering to patients,severely affecting their daily lives,and its suddenness is one of the main reasons for disability and even death in epilepsy patients.Neurologists monitor and diagnose patients using electroencephalograms(EEGs),but there are issues of timeconsuming,subjective interpretation,and high error rate.Therefore,with the rapid development of signal processing technology and artificial intelligence,developing computer-aided automatic epileptic seizure detection and prediction methods has promising and important significance.Automatic epileptic seizure detection can provide relatively objective reference for clinical diagnosis and evaluation,effectively reduce the workload of doctors,and improve diagnostic efficiency.This thesis proposes a single-channel epileptic seizure detection algorithm based on EEG power spectral parameterization analysis.The method extracts key features related to epileptic seizure EEG signals based on the power spectral density(PSD)parameterization analysis model,which match the description of epileptic seizure mechanism by epileptologists.Then,statistical analysis is used for channel selection to screen the most relevant channel for each seizure.Finally,combined with support vector machine classifier,a single-channel,low-power and superior performance epileptic seizure detection is achieved.The proposed algorithm obtains an average sensitivity,specificity,accuracy,precision and F1 score of 95.6%,99.2%,98.6%,95.5%and 95.5%,respectively,on the CHBMIT dataset.The experimental results show that the proposed method performs better than other single-channel methods,is comparable in performance to dualchannel detection algorithms,and has reliable clinical application value.The epileptic seizures prediction not only assists clinical doctors in diagnosis,but also enables early intervention to alleviate the physical and mental burden of patients and improve their quality of life.Therefore,this thesis improves the prediction performance from three aspects:time-frequency analysis of EEG signals,stability of unsupervised feature learning models,and design of back-end classifiers.Combined with a channel selection strategy based on PSD parameterization analysis,a semi-supervised epileptic seizure prediction model(ST-WGAN-GP-Bi-LSTM-CS)is proposed to improve the prediction performance.The ST-WGAN-GP-Bi-LSTMCS prediction model proposed in this thesis was verified on the CHB-MIT scalp EEG dataset,achieving AUC,sensitivity,and specificity indicators of 97.98%,91.42%and 96.09%,respectively.Compared with existing semi-supervised methods,this method improved the original performance indicators by 25.35%,29.21%and 63.06%,and performed equally well as the supervised prediction model based on CNN,using only 4 channels.This method effectively improves the prediction performance of semi-supervised deep learning models and plays an optimized role in unsupervised feature extraction in epileptic seizure prediction.The research findings of this article indicate that computer-aided detection and prediction of epileptic seizures based on EEG signals can improve performance by optimizing feature extraction,adaptive channel selection,reducing label data dependence,and establishing interpretable models,which can contribute to the development of epileptology research.