Research On Time Series Medical Data Analysis Method Based On LSTM

Recently,deep learning has been widely used in the medical field.Risk profile model based on time series medical data is a very popular research today.Effective use of medical data for medical risk profile is of great significance for early diagnosis and prevention.Long Short-Term Memory network has memory,so it has certain advantages in learning the nonlinear features of sequences,which is widely used in natural language processing and various time series prediction.Under the quick pace of people’s life,cardiovascular diseases(CVD)have become one of the major killers threatening people’s life and health.Arrhythmia is a kind of CVD,which is due to abnormal sinus node excitation or excitation occurring outside sinus node.The conduction of excitation is slow,blocked or conducted through abnormal channels,i.e.the origin of cardiac activity and conduction disorder lead to abnormal frequency or rhythm of cardiac beat.It is a relatively common cardiac disease clinically,which can seriously lead to stroke and even sudden cardiac death.ECG data records the changes of electrical activity generated by each heartbeat cycle of the heart,which is a curve of voltage change with time and has strong timing.Electrocardiogram(ECG)is an important basis for doctors to diagnose cardiac diseases.It is very important to use ECG to correctly diagnose arrhythmia and prevent it early.Traditional ECG diagnosis mainly depends on doctors reading ECG,which takes a long time.This puts forward a very high requirement for the doctor’s diagnosis level and experience,which may lead to the doctor not being able to make a diagnosis of the patient’s condition in time,thus delaying the treatment.With the rapid development of computer technology,computer-aided arrhythmia diagnosis has arisen,including ECG preprocessing,ECG feature extraction,arrhythmia diagnosis based on intelligent algorithms,etc.ECG data is a typical time series data in the medical field.Therefore,this thesis makes use of its chronology and studies the classification of arrhythmia based on LSTM.The main contents are as follows:1.A classification model of arrhythmia based on LSTM was constructed in this thesis.Five kinds of beats in MIT-BIH arrhythmia database,N(normal beat),V(ectopic ventricular beat),L(left bundle branch block),R(right bundle branch block)and A(atrial premature beat),were classified.This method has obvious advantages based on varification by comparing with SVM,RNN,BP and KNN classification models.2.Under LSTM model and SVM model,the influence of different ECG data segment lengths on arrhythmia classification was studied.The experimental results showed that different ECG data segment lengths perform generally under SVM model,and the overall classification effect is not as good as LSTM model,which is very unstable.Under LSTM model,when the ECG data segment was 400 to 500,the classification accuracy was higher;When the data segment length was 450,the classification accuracy was the best,and when the data segment length was greater than600,the classification accuracy started to decline.