Classification Of Arrhythmia Uisng Combination Of CNN And LSTM Techniques

Arrhythmia refers to excessive,slow or irregular heart rate caused by abnormal origin or frequency of excitation,velocity of conduction or any link of pathway.Arrhythmia may be caused by impulse formation or disturbance of physiological conduction,or both,but it is not always an irregular heart activity.Arrhythmias can occur in healthy hearts with little impact(e.g.respiratory sinus arrhythmias,a natural periodic change of heart rate corresponding to respiratory activity),but they may also indicate a serious problem that may lead to stroke or sudden cardiac death.Therefore,the automatic detection and classification of arrhythmia is very important in clinical cardiology,especially for real-time detection tasks.This is achieved by analyzing ECG and its extracted features.ECG signals are easily affected by the outside and the human body,and because of the great individual differences of patients,the traditional method of analysing ECG by experts is prone to subjectivity,which leads to misdiagnosis.Traditional machine learning methods rely on prior knowledge,and need to design and extract features of ECG signals.It is difficult to mine the deep features behind massive ECG signals.In recent years,deep learning has developed rapidly,and shown great performance in many fields.In this thesis,a model based on the combination of Long Short-Term Memory(LSTM)and Convolutional Neural Networks(CNN)is proposed to implement the classification of arrhythmias.The main work of this thsis includes:1.A method of arrhythmia classification based on Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)is proposed.The automatic classification of five arrhythmias,N,S,V,F and Q,recommended by AAMI,was realized.Two different schemes were designed to optimize the model.The feasibility of the optimization schemes were verified.2.By comparing the sensitivity and positive predictivity rate of S and V type beats under different data segment lengths,it can be concluded that the length of the input ECG data segment had an important influence on the classification performance of the network.When the length of data segment was between 400 and 550 sampling points,the network model can get better classification results.The influence of signal slicing on experimental results was discussed.Although the results were not as good as segmentation scheme,this method is closer to practical application,and makes the signal diagnosis process simpler and more general.3.MIT-BIH standard database was used to verify the classification effect of the model.The experimental results show that the sensitivity,specificity and overall accuracy of the model are 95.01%,98.58% and 97.73%,respectively.In conclusion,the method of arrhythmia classification based on Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)proposed in this paper can automatically learn the deep features of the input ECG signal,thus implement the classification of arrhythmia,which have a certain positive significance for the relevant clinical application.