基于滤波重构和卷积神经网络的心电信号分类

doi:10.16180/j.cnki.issn1007-7820.2019.11.002

摘要/Abstract

摘要：

计算机自动分类心电信号能够减轻医生工作压力并大幅提高诊断速度和准确率。文中针对传统算法中特征提取过程复杂及抗干扰能力弱的问题,提出了一种结合滤波重构和卷积神经网络的心电信号分类算法。该算法首先通过传统信号滤波和心拍序列重构去除原始心电信号中的噪声干扰,然后构建卷积神经网络来自动学习心电信号特征并完成分类。在PhysioNet/CinC Challenge 2017数据集上的分类实验结果表明,该方法的平均F₁(查准率、召回率的调和平均)达到了0.8471,优于人工特征提取和常规卷积网络方法,且具有很强的抗干扰能力。

关键词: 卷积神经网络, 心电信号, 特征自动提取, 序列重构, 信号滤波, 分类算法

Abstract:

Automatic classification of ECG signals by computers can relieve work pressure of doctors and greatly improve diagnosis speed and accuracy. Aiming at the problem of complex feature extraction process and weak anti-interference ability in traditional algorithms, this paper proposed an ECG signal classification algorithm combined with filtering-reconstruction and convolutional neural networks. Firstly, the traditional signal filtering and heartbeat sequence reconstruction were used to remove the noise interference in the original ECG signal, and then the convolutional neural network was constructed to automatically learn the ECG signal feature and completed the classification. The results of the classification experiments on the PhysioNet/CinC Challenge 2017 dataset showed that this method had an average F₁ (the average of the precision and recall ratio) of 0.8471, which was better than the methods based on artificial feature extraction and conventional convolutional network, and had strong anti-interference ability.

Key words: convolutional neural network, electrocardiogram signal, automatic feature extraction, sequence reconstruction, signal filtering, classification algorithm

中图分类号:

TP391

韦张跃昊,钱升谊. 基于滤波重构和卷积神经网络的心电信号分类[J]. 电子科技, 2019, 32(11): 7-11.

WEI Zhangyuehao,QIAN Shengyi. ECG Signal Classification Based on Filtering-Reconstruction and Convolutional Neural Network[J]. Electronic Science and Technology, 2019, 32(11): 7-11.

图/表 7

图1

图2

图3

图4

图5

表1

图6

参考文献 19

[1]	杨雪梅 . 动态心电图与常规心电图诊断冠心病患者心肌缺血及心律失常的临床效果对比分析[J].中国现代医生, 2017(7):85-87.
	Yang Xuemei . Comparison and analysis of clinical effect of dynamic electrocardiogram and conventional electrocardiogram in the diagnosis of myocardial ischemia and arrhythmia in the patients with coronary heart disease[J].China Modern Doctor, 2017(7):85-87.
[2]	Tateno K, Glass L . Automatic detection of atrial fibrillation using the coefficient of variation and density histograms of RR and deltaRR intervals[J]. Medical & Biological Engineering & Computing, 2001,39(6):664-671.
[3]	宋莉, 孟庆建, 张光玉 , 等. 基于波形特征和SVM的心电信号自动分类方法研究[J].中国医学物理学杂志 2010(4):2043-2046.
	Song Li, Meng Qingjian, Zhang Guangyu , et al. Methods study of automatic classifying ECG signals based on wave form features and SVM[J].Chinese Journal of Medical Physics, 2010(4):2043-2046.
[4]	Elhaj F A, Salim N, Harris A R , et al. Arrhythmia recognition and classification using combined linear and nonlinear features of ECG signals[J]. Computer Methods & Programs in Biomedicine, 2016,127(C):52-63.
[5]	Acharya U R, Fujita H, Lih O S , et al. Automated detection of arrhythmias using different intervals of tachycardia ECG segments with convolutional neural network[J]. Information Sciences 2017(3):405-411.
[6]	黄政钦, 孙静, 张丽娜 , 等. 心音、心电采集系统设计与信号预处理[J]. 电子测量技术, 2014,37(9):117-121.
	Huang Zhenqin, Sub Jing, Zhang Lina , et al. ECG and heart sound acquisition system and signal preprocessin[J]. Electronic Measurement Technology, 2014,37(9):117-121.
[7]	庞宇, 邓璐, 林金朝 , 等. 基于形态滤波的心电信号去除基线漂移方法[J]. 物理学报, 2014,63(9):428-433.
	Pang Yu, Deng Lu, Lin Jinzhao , et al. A method of removing baseline drift in ECG signal based on morphological filtering[J]. Acta Physica Sinica, 2014,63(9):428-433.
[8]	周静 . 心电信号中工频干扰的消除[J].生物医学工程研究 2003(4):61-64.
	Zhou Jing . Elimination of power-line interference from ECG signals[J].Journal of Biomedical Engineering Research, 2003(4):61-64.
[9]	江伟, 袁芳, 杨柳青 . 心电信号中去除肌电干扰信号的研究[J]. 安徽大学学报(自然科学版), 2017,41(3):85-89.
	Jiangh Wei, Yuan Fang, Yang Liuqing . Study of removing EMG interference signal for ECG signal[J]. Journal of Anhui University(Natural Sciences), 2017,41(3):85-89.
[10]	李楠, 王聚河, 黄文松 . 基于改进Morlet小波的超声回波信号处理[J]. 现代电子技术, 2012,35(21):49-51.
	Li Nan, Wang Juhe, Huang Wensong . Research on ultrasonic signal processing based on improved Morlet wavelet for felted composite[J]. Modern Electronics Technique, 2012,35(21):49-51.
[11]	朱先和, 杨世平 . 基于小波变换对信号噪声的处理研究[J]. 电子科技, 2016,29(6):128-131.
	Zhu Xianhe, Yang Shiping . Signal de-noising research based on wavelet transform[J]. Electronic Science and Technology, 2016,29(6):128-131.
[12]	Martinez J P, Almeida R, Olmos S , et al. A wavelet-based ECG delineator: evaluation on standard databases[J]. IEEE Transactions on Bio-medical Engineering, 2004,51(4):570-579.
[13]	Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks [C].Doha:International Conference on Neural Information Processing Systems, 2012.
[14]	Clifford G, Liu C, Moody B, et al. AF classification from a short single lead ECG recording [C].Rennes:Computing in Cardiology Conference, 2017.
[15]	Hornero R . Optimal parameters study for sample entropy-based atrial fibrillation organization analysis[J]. Comput Methods Programs Biomed, 2010,99(1):124-132.
[16]	Antink C H, Leonhardt S, Walter M. Fusing QRS detection, waveform features, and robust interval estimation with a random forest to classify atrial fibrillation [C].Rennes: Computing in Cardiology Conference, 2017.
[17]	Datta S, Puri C, Mukherjee A, et al. Identifying normal, AF and other abnormal ECG rhythms using a cascaded binary classifier [C].Rennes:Computing in Cardiology Conference, 2017.
[18]	Xiong Z, Stiles M, Zhao J. Robust ECG signal classification for the detection of atrial fibrillation using novel neural networks [C].Rennes:Computing in Cardiology Conference, 2017.
[19]	Liu C, Li Q, Suresha P B, et al. Combining multi-source features and support vector machine for heart rhythm classification [C].Rennes:Computing in Cardiology Conference, 2017.

模型方法	F₁_n	F_1a	F_1o	F_1P	F₁
文献一	0.896 3	0.753 4	0.722 2	0.624 9	0.790 6
文献二	0.880 0	0.770 0	0.720 0	-	0.790 0
文献三	0.909 5	0.797 8	0.771 9	0.826 4	0.826 4
文献四	0.900 0	0.820 0	0.750 0	-	0.820 0
文献五	0.921 0	0.835 0	0.739 0	0.870 0	0.832 0
本文	0.921 6	0.824 5	0.795 3	0.623 0	0.847 1