Research On Arrhythmia Classification And ECG Identification Based On BiLSTM

The number of people with cardiovascular disease have been increased year by year in China and the world.There are corresponding arrhythmias before the onset of cardiovascular disease,or as a complication of cardiovascular disease,therefore,arrhythmia plays a key role in cardiovascular disease of prevention,diagnosis and treatment.Arrhythmia has the characteristics of suddenness and timeliness of rescue,which has caused the acquisition equipment to fail to collect arrhythmias in time.24-hour acquisition equipment has brought tens of thousands of heart beat cycle data,the traditional artificial recognition of ECG is easy to cause misdiagnosis.The computer is particularly important as an auxiliary diagnostic method for arrhythmia.As a new biometrics technology,ECG signal has the advantages of difficult to forge,low cost and high stability.Traditional filtering method is only considered the forward filter of the signal;Most of the current arrhythmia classification algorithms have achieved good results based on balanced distribution of heartbeat data set,and have relatively poor results on data sets with unbalanced distributed heartbeats.Traditional ECG identification model is only considered healthy or non-healthy individual,the model generation is not strong.Based on public data sets,filtering of ECG,arrhythmia classification and ECG identification has been deeply studied in this thesis.The main work of this thesis includes:Firstly,the EKS(Extended Kalman Smoother)filtering algorithm has been used to denoise the ECG signal,it includes forward extended Kalman filtering and reverse smoothing filtering,the end point of the forward extended Kalman filter is taken as the start point of the reverse smoothing filter.EKS compares the minimum mean square error and signal-to-noise ratio of the filtered signal with wavelet soft threshold filtering,median filtering and extended Kalman filtering,EKS effectively improves the signal-to-noise ratio of the signal and reduces the minimum mean square error of the signal.Secondly,a method of classification arrhythmia using the model of Bi LSTM(Bidirectional Long Short-Term Memory Networks)was proposed.The five types of arrhythmia classification results of five classification models BP?SVM?1D-CNN?LSTM and Bi LSTM were compared and analyzed;the overall accuracy BP is 97.90%,the overall accuracy SVM is 98.71%,the overall accuracy LSTM is 98.75%,the overall Accuracy Bi LSTM is 98.75%.Thirdly,a method of ECG identification using the Bi LSTM was proposed.Three datasets were composed of selected data from two public ECG datasets.The classification accuracy rate of heart beat classification for three data sets is over 96%,and the accuracy rate of identity recognition is 100%.