Epilepsy Warning Based On Machine Learning

Epileptic seizures are characterized by sudden and repetitive,and the high risk of seizures leads to epilepsy,which has a significant impact on the daily work and life of patients.Therefore,the realization of early warning of epilepsy has important clinical research significance.In this paper,the EEG signals of 15 minutes before seizures are classified and judged,and when the EEG signals are found to belong to this period,the model warns to remind the medical staff to give medication or other treatment to the patients.(1)This paper adopts the method of medical diagnosis of epilepsy.That is to say,EEG signals are used to judge the early warning of epilepsy.In this paper,ICA-EMD denoising method is designed in the denoising stage.In the analysis of epileptic signals,firstly,according to the general characteristics of EEG signals,denoising processing is carried out,and ICA fast algorithm is used to remove most of the artifact noise.In view of the high frequency noise produced by the machine in the process of EEG acquisition,this paper selects the method of EMD decomposition and reconstruction to reduce the noise of EEG signal.(2)The DWPT-SVM model is designed for epilepsy warning,the wavelet packet transform is used to extract the energy spectral density,and then the PCA dimension reduction is used to obtain the 15-dimensional feature vector,which is combined with Shannon entropy,logarithmic energy entropy and characteristic norm energy entropy to form the feature vector.In the design of SVM classifier,the grid optimization method is used to determine the parameters,and in the experiment,the cross-validation method is used to model and evaluate the epilepsy warning model.The experiments show that the effectiveness of the model is not sensitive to the length of the window when extracting features.Considering the real-time requirement of epileptic seizure prediction,the 1s window is selected to build the model features.The average accuracy of the final DWPTSVM model is 90.674%,and the average sensitivity is 90.255%.However,the model fluctuates greatly in the early warning performance of different patients.Not stable enough.(3)In this paper,a neural network architecture based on bi-layer Bi-LSTM is proposed to predict epilepsy.The double-layer Bi-LSTM model inputs the preprocessed timing signals.The improved bi-layer Bi-LSTM model is trained by Adam optimization,and achieves 92.714% and 92.396% results in average accuracy and average sensitivity,respectively.Compared with the other three network models,the superiority of the improved double-layer Bi-LSTM can be determined.Finally,after training,the data of all patients were processed by sliding average method,and only a few patients were misreported,and there was no underreport.To sum up,this paper applies machine learning and deep learning technology to study the field of epilepsy early warning.Through the comparison of model experiments of all patients,it is found that the double-layer Bi-LSTM model has good performance in accuracy and model stability of different patients,which adds a certain theoretical basis for the practical application of epilepsy early warning in clinical practice.