删除或更新信息,请邮件至freekaoyan#163.com(#换成@)

基于卷积神经网络的第一导联心电图心拍分类

本站小编 Free考研考试/2022-01-16

庞彦伟, 李潇, 梁金升, 何宇清
AuthorsHTML:庞彦伟, 李潇, 梁金升, 何宇清
AuthorsListE:Pang Yanwei, Li Xiao, Liang Jinsheng, He Yuqing
AuthorsHTMLE:Pang Yanwei, Li Xiao, Liang Jinsheng, He Yuqing
Unit:天津大学电气自动化与信息工程学院,天津 300072
Unit_EngLish:School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
Abstract_Chinese:第一导联心电图心拍的分类具有重要的医学价值, 可以用来判断心脏的健康状况.采用深度卷积神经网络的方法, 设计了针对单导联心电图这种特殊一维信号的卷积神经网络.该卷积神经网络具有层数多、卷积核尺度多样、参数量小等特点, 能有效对第一导联心电图心拍进行分类.首先将心电数据进行预处理输入网络, 经过一系列卷积、池化操作, 最终输出分类结果.将该网络应用于INCART数据库, 对超过17×104条第一导联心电图数据进行分类实验, 取得了98% 的准确率、90% 的敏感度和86% 的阳性预测值的分类性能.实验结果表明, 所采用的方法可以对第一导联心电图心拍进行很好的分类, 并可应用于可穿戴设备和远程监护领域.
Abstract_English:The classification of first lead electrocardiogram(ECG)heartbeats has significant medical value. It can be used to diagnose the health of heart. A deep convolutional neural network(CNN)for single lead ECG,a special one-dimensional signal,was proposed in this paper. The proposed CNN was characterized by a very deep structure,multi-scale convolution kernels,and meanwhile a small parameter size. The proposed method can classify the first lead ECG heartbeats effectively. Firstly,preprocessed ECG data was imported into the network. Then,the classification results were exported through a series of convolution and pooling operation. This CNN was applied to the St. Petersburg Institute of Cardiological Technics 12-lead Arrhythmia Database(INCART). The classification experiments were carried out with more than 170 thousand first lead ECG data. The results obtained were: accuracy 98% ,sensitivity 90% and positive predictive value 86% . Experiment results show the proposed method can make a good classification of first lead ECG heartbeats. It can further apply in wearable devices and remote monitoring areas.
Keyword_Chinese:第一导联; 心电图; 卷积神经网络; 可穿戴设备; 远程监护
Keywords_English:first lead; electrocardiogram; convolutional neural networks; wearable devices; remote monitoring

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6088
相关话题/卷积 神经网络