Research And Implementation Of Abnormal ECG Classification Based On Deep Learning

The human body's ECG signal can reflect the relationship between cardiac health and cardiovascular-related diseases.It is widely used in clinical diagnosis or prevention of diseases.Because of individual differences,the performance of ECG signals reflecting the same kind of heart disease may vary in different patients.This difference often requires experienced doctors to make accurate judgments.Under the premise of lack of medical resources and lack of professional doctors' support,it will lead to misdiagnosis and missed diagnosis.With the development of information technology,computer aided diagnosis technology based on artificial intelligence can automatically classify and predict cardiovascular diseases according to the electrocardiogram signals of patients,which can significantly improve the efficiency of diagnosis.Therefore,searching for a method to improve the classification accuracy of ECG signals has always been the goal of biomedical engineers.In recent years,the classification algorithm of abnormal ECG based on machine learning method has improved the accuracy of classification of abnormal ECG signals,but the traditional machine learning framework needs to calibrate the characteristics of abnormal ECG signals first.This method that relies heavily on artificial features is difficult to identify subtle abnormal information,resulting in misjudgment of diagnosis results.The classification method of abnormal ECG signals based on deep learning can automatically mine the hidden features of data and realize the automatic classification of abnormal ECG signals.Based on the Long Short-Term Memory Network,this paper uses a classification method of Bidirectional Long Short-Term Memory Network,which can accurately classify the electrocardiogram signals of N,S,V and F,and achieve a total accuracy of 98.20%,which is 0.38% and 1.58% higher than the OneDimensional Convolutional Neural Network and the Long Short-Term Memory Network respectively.The main research contents and work of this paper include:1)ECG signal preprocessing.The MIT-BIH database was divided according to the arrhythmia classification criteria established by AAMI(Association for the Advancement of Medical Instrumentation),then the ECG signals were de-noised,and QRS positioning,and beats interception were performed to provide appropriate data for the model.2)ECG anomaly classification based on Long Short-Term Memory Network(BiLSTM).The total accuracy,accuracy,precision,recall rate and specificity derived from the confusion matrix were compared with the classification effects of BiLSTM and 1D-CNN and LSTM.The results showed that BiLSTM model was better than the other two models in the classification of ECG anomalies.To sum up,BiLSTM model is used to classify ECG anomalies in this paper,which can autonomously mine the deep features of ECG data with high accuracy,and has certain positive significance for assisting physicians to judge ECG abnormalities.