基于深度残差记忆网络的心律失常分类方法  

Classification Method of Arrhythmia Based on Deep Residual Memory Network

作　　者：沈尧 马小虎[1] 张扬 杨东东 SHEN Yao;MA Xiao-hu;ZHANG Yang;YANG Dong-dong(School of Computer Science and Technology,Soochow University,Suzhou Jiangsu 215000,China)

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215000

出　　处：《计算机仿真》2022年第12期342-346,518,共6页Computer Simulation

基　　金：江苏省自然科学基金(BK20141195);江苏省高校优势学科建设工程。

摘　　要：心律失常是一种严重危害人类健康的心血管疾病,其主要的确诊手段靠心电图(ECG)。采用计算机辅助心电信号实现心律失常自动分类可有效避免人工误差,提高诊断效率并降低成本。基于心电信号,提出一种改进的深度残差网络(ResNet)方法,对美国医疗仪器促进协会(Association for Advancement of Medical Instrumentation, AAMI)建议的五种心律失常实现自动分类。上述方法首先通过加入CBAM机制的ResNet提取心电信号的重要特征。随后进一步结合长短时记忆网络(LSTM)构建L-ResNet模型。经以麻省理工心律失常数据库(MIT-BIH)为数据集进行验证,得出99.51%的分类准确率,并在灵敏度等其它性能指标上均有提升,最高可达99.98%。Arrhythmia is a cardiovascular disease that seriously endangers human health. The main diagnosis method is electrocardiogram(ECG). The use of computer-assisted ECG signals to realize automatic classification of arrhythmias can effectively avoid manual errors, improve diagnosis efficiency and reduce costs. To this end, this paper proposes an improved deep residual network(ResNet) method based on the ECG signal to automatically classify the five arrhythmias proposed by the Association for Advancement of Medical Instrumentation(AAMI). First, the important features of the ECG signal through ResNet with CBAM mechanism were extracted. Additionally, the Long Short Term Memory network(LSTM) was further combined to construct the L-ResNet model. The MIT arrhythmia database(MIT-BIH) was used as the data set to verify that the classification accuracy is 99.51%,and other performance indicators such as sensitivity are improved, up to 99.98%.

关 键 词：深度残差网络 长短时记忆网络 心律失常 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]