基于空洞卷积神经网络的心律失常分类算法  

Arrhythmia detection algorithm based on dilated convolutional neural network

作　　者：秦默然 李宙童 翟月英 史纪广 纪洁维 常胜[1] 王豪[1] 何进[1] 黄启俊[1] QIN Moran;LI Zhoutong;ZHAI Yueying;SHI Jiguang;JI Jiewei;CHANG Sheng;WANG Hao;HE Jin;HUANG Qijun(School of Physics and Technology,Wuhan University,Wuhan 430072,China;Department of Cardiology,Huangpu Brunch of Shanghai Ninth People's Hospital,Shanghai Jiaotong University School of Medicine,Shanghai 200011,China;Department of Electronic Information Engineering,Wuhan Qingchuan University,Wuhan 430204,China)

机构地区：[1]武汉大学物理科学与技术学院,湖北武汉430072 [2]上海交通大学医学院附属第九人民医院黄浦分院心血管内科,上海200011 [3]武汉晴川学院电子信息工程系,湖北武汉430204

出　　处：《中国医学物理学杂志》2023年第1期87-94,共8页Chinese Journal of Medical Physics

基　　金：国家自然科学基金(81971702,61874079,61774113)。

摘　　要：本文提出了一种基于卷积网络的心电信号分类算法,设计了空洞卷积池化金字塔模块,通过不同尺寸的空洞卷积提取信息,再将各通道的信息聚合,在增强网络的特征提取能力的同时可以降低参数量。本文聚焦于窦性心律、房性早搏、心动过速以及心动过缓4种分类,使用的心电图数据集来自医院的实测数据,数据集包含75000名不同检测者的心电记录。经过测试,本文提出的模型在该数据集上取得了0.89的F1值,另外在CinC2017数据集上也达到了0.87的F1值。实验结果表明该分类算法具有优秀的特征提取和分类能力,在心电信号的实时分类中具备应用前景。An electrocardiogram(ECG)signal classification algorithm based on convolutional network is proposed.The algorithm adopts an atrous spatial pooling pyramid module to extract information through atrous convolution of different sizes,and aggregates the information of each channel for enhancing the ability of feature extraction and reducing the number of parameters.The study focuses on the categories of sinus rhythm,premature atrial contraction,tachycardia and bradycardia,and a real ECG data set from a hospital which contains ECG records of 75000 different subjects is used for experiment.The results reveal that the proposed model reaches an F1 score of 0.89 on the real ECG data set,and also achieved an F1 score of 0.87 on the CinC2017 data set,which indicates that the classification algorithm has excellent feature extraction and classification capabilities,and has application prospects in the real-time classification of ECG signals.

关 键 词：心律不齐 神经网络 心电图 深度学习 

分 类 号：R318[医药卫生—生物医学工程]