Classification Method And Interpretability Of Arrhythmia Based On Deep Learning

With the improvement of people’s living standard,heart disease has become one of the common diseases threatening human health.Electrocardiogram is an effective means to diagnose cardiovascular diseases.With the rapid development of electrocardiography and the shortage of cardiologists,how to accurately and automatically monitor arrhythmia under the condition of long-term electrocardiography monitoring has become a research hotspot.In order to improve the accuracy and interpretability of the classification model,some researches are investigated to implement classification of arrhythmia.In addition,an arrhythmia classification system is developed to simply and stably monitor the ECG status of users.The specific research contents of this paper are listed as follows.(1)A hybrid 1D Resnet-GRU method consisting of Resnet and gated relapse Unit(GRU)modules is proposed for the classification of arrhythmias from 12-lead ECG recordings.In addition,Focal Loss function was used to solve the problem of unbalance data set.Based on the proposed 1D Resnet-GRU model,the Grad-CAM++ mechanism was applied to the trained network model to generate thermal images superimposed on the original signals,which provided interpretable basis for the classification results of arrhythmia classification model.The experimental results showed that the proposed method was effective in classifying 9 arrhythmias,and the accuracy of classification was improved.In addition,Grad-CAM ++ provides classification based on models consistent with clinical diagnostic methods for arrhythmias.(2)Based on the Squeeze and Excitation Block(SE Block),a simple and effective method is designed to improve the arrhythmias classification.In this method,the spatial characteristics of ECG were extracted by residual network,and the temporal characteristics of ECG were extracted by cyclic network module.Finally,the compression excitation module was used to improve the classification effect.The ECG datasets were measured from the second affiliated hospital of Zhejiang University,which includes seven different ECG types.Ablation and comparison experiments were also conducted to validate the performances of the proposed model.The experimental results show that the proposed method is simple and effective,and can maintain the high performance of arrhythmia classification,and has good applicability in practical application scenarios.(3)Android Studio software was used to develop an arrhythmia monitoring system with real-time monitoring,intelligent diagnosis and historical rollback functions based on MVP mode.The system can connect with ECG monitoring devices through Bluetooth,collect ECG in real time,and return diagnostic results from cloud server to client,thus achieving early warning function.Meanwhile,the deep learning model in the cloud can be flexibly changed according to the requirements of different ECG classification methods to achieve ECG classification results that meet the needs of clinical diagnosis.Finally,the cloud deep learning model can be iteratively updated through the collected ECG data to enhance the classification effect.This paper mainly studies the classification algorithm of arrhythmia,the interpretability of classification results and the realization of arrhythmia diagnosis system.The research in this paper can effectively improve the accuracy of arrhythmia classification,and provide the interpretability of classification results through Grad-CAM++ generating thermal imaging,so as to provide decision-making basis for the clinical diagnosis of arrhythmia.It is of great guiding value to medical clinical practice.