Research Of Arrhythmia Classification Based On LSTM

Cardiovascular disease,also known as the circulatory system disease,is characterized by high morbidity,high disability and high mortality.It is known as the "biggest killer" that endangers human health.Arrhythmia plays an important role in cardiovascular disease.It can lead to sudden death and also lead to heart failure.ECG(Electrocardiogram)signals are a series of bioelectrical signals that reflect heart activities and an important indicator of the diagnosis of arrhythmias.In clinical diagnosis,the observation of ECG signals is mainly achieved through ECG.Since the ECG signal has characteristics such as small amplitude and low frequency,it is easily affected by the outside world and the human body itself.In addition,due to the large individual differences in patients,a wide variety of arrhythmias,and the subjectivity of doctors in interpreting ECG,traditional manual analysis of ECG signals by professional doctors is time-consuming and laborious,and it is easy to cause misdiagnosis.A reliable,accurate and rapid automatic classification method of arrhythmia not only can greatly reduce the workload of doctors,but also can improve the efficiency and accuracy of diagnosis of arrhythmias,and has high practical value and social value.The arrhythmia classification based on LSTM(Long Short-Term Memory,LSTM)has been deeply studied in this thesis.The main work of this thesis includes:1.A method of automatic classification of arrhythmias using the deep learning model LSTM was proposed,which realized the automatic classification of 5 types of arrhythmias suggested by AAMI(Association for Advancement of Medical Instrumentation): N,S,V,F and Q.By using MITBIH Arrhythmia Database,the classification results of five classification models of LSTM,SVM,KNN,RF and BP were compared and analyzed.The feasibility and advantages of utilizing LSTM were verified.2.The effect of different ECG data segment lengths on arrhythmia classification results based on LSTM is studied.When the ECG data segment length is between 400 and 600 sampling points,the arrhythmia classification results of LSTM are better.Especially for 550 sampling points,the LSTM based arrhythmia classification achieves the best performance.When the number of sampling points is too many,the classification performance decreases rapidly.3.By comparing the five types of arrhythmias classification results of LSTM and SVM,KNN,RF and BP on different length of the ECG data segments,it is found that when the length of the ECG data segment is between 350 and 650 sampling points,LSTM has obvious advantages in the classification accuracy.The arrhythmia classification performance of LSTM is greatly affected by the length of the ECG data segment,and the performance of other four algorithms is less affected by the length of the ECG data segment.In conclusion,the proposed arrhythmia classification algorithm based on LSTM can extract the deep features from ECG,and then effectively realize automatic classification of arrhythmia.It would have certain positive significance for related applications in clinic.