Please wait a minute...
浙江大学学报(工学版)
浙江大学学报(工学版)  2020, Vol. 54 Issue (5): 1039-1048    DOI: 10.3785/j.issn.1008-973X.2020.05.023
计算机技术     
基于模式识别和集成CNN-LSTM的阵发性房颤预测模型
杨萍(),王丹*(),康子健,李童,付利华,余悦任
北京工业大学 信息学部，北京 100124
Prediction model of paroxysmal atrial fibrillation based on pattern recognition and ensemble CNN-LSTM
Ping YANG(),Dan WANG*(),Zi-jian KAGN,Tong LI,Li-hua FU,Yue-ren YU
Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China
 全文: PDF(1051 KB)   HTML
摘要：

为了适用于长期心电监护和ICU等实时性、数据密集型应用场合，提出可在8 Hz采样频率的1 min心电图（ECG）片段上提前45 min预测阵发性房颤（PAF）发作的实时预测模型. 采用概率符号化模式识别方法，在降采样后的ECG序列上提取出1 min窗口内的模式转移特征，降低模型的计算量和对存储空间的需求，确保实时预测的效果. 提出卷积神经网络（CNN）和长短-期记忆网络（LSTM）的混合模型（CNN-LSTM），用于提取模式转移特征内隐含的局部空间特征和时间依赖特征. 为了提升模型泛化能力，构建基于CNN-LSTM的集成分类器. 采用Spark Streaming技术完成对ECG流式数据的读、写和计算，实现数据和模型之间的低延迟通信. 所提模型在公开数据集上的准确率、灵敏度和特异度分别为91.26%、82.21%、95.79%. 模型处理总延迟平均为2 s，满足实时PAF预测需求.
关键词： 阵发性房颤;  心电图（ECG）;  预测;  概率符号化模式识别;  卷积神经网络（CNN）;  长短-期记忆网络（LSTM）;  Spark Streaming    
Abstract:

The real-time prediction model that can predict the onset of paroxysmal atrial fibrillation (PAF) 45 min in advance on the one minute electrocardiogram (ECG) segment with 8 Hz sampling frequency was proposed, for real-time and data-intensive application scenarios such as long-term ECG monitoring and intensive care units (ICU). The probabilistic symbolic pattern recognition method was used to extract the pattern transition features within one minute window of down sampled ECG sequence, reducing the calculation complexity of the model and the demand for storage space, so as to ensure the effect of real-time prediction. A hybrid model (CNN-LSTM) of the convolutional neural network (CNN) and the long short-term memory (LSTM) was proposed to extract local spatial features and time-dependent features implied in pattern transition features. An ensemble classifier based on CNN-LSTM was constructed to improve the generalization ability of the model. Spark Streaming technology was used to read, write and calculate ECG streaming data, and low latency communication between data and model was realized. The accuracy, sensitivity, and specificity of the proposed model were 91.26%, 82.21%, and 95.79% respectively. The average delay of model processing was 2 s, which can meet the real-time PAF prediction demand.
Key words: paroxysmal atrial fibrillation (PAF)    electrocardiogram (ECG)    prediction    probabilistic symbolic pattern recognition    convolutional neural network (CNN)    long short-term memory (LSTM)    Spark Streaming
收稿日期: 2019-11-27 出版日期: 2020-05-05
CLC:  TP 183  
基金资助: 国家自然科学基金资助项目（61672506）；国家自然科学基金青年基金资助项目（61902010）
通讯作者: 王丹     E-mail: yangping_sx@163.com;neuwd@sina.com.cn
作者简介: 杨萍（1987—），女，博士生，从事机器学习、大数据和生命健康领域研究. orcid.org/0000-0002-7091-2289. E-mail： yangping_sx@163.com
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章　　
杨萍
王丹
康子健
李童
付利华
余悦任

引用本文:

杨萍,王丹,康子健,李童,付利华,余悦任. 基于模式识别和集成CNN-LSTM的阵发性房颤预测模型[J]. 浙江大学学报(工学版), 2020, 54(5): 1039-1048.

Ping YANG,Dan WANG,Zi-jian KAGN,Tong LI,Li-hua FU,Yue-ren YU. Prediction model of paroxysmal atrial fibrillation based on pattern recognition and ensemble CNN-LSTM. Journal of ZheJiang University (Engineering Science), 2020, 54(5): 1039-1048.

链接本文:

http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2020.05.023    或    http://www.zjujournals.com/eng/CN/Y2020/V54/I5/1039

模式 PTF2 PTP2
a b c 总计 a b c
aa 0.0 0.0 0.0 0.0 0.0 0.0 0.0
ab 0.0 0.0 1.0 1.0 0.0 0.0 1.0
ac 0.0 1.0 1.0 2.0 0.0 0.5 0.5
ba 0.0 0.0 2.0 2.0 0.0 0.0 1.0
bb 0.0 0.0 0.0 0.0 0.0 0.0 0.0
bc 0.0 0.0 1.0 1.0 0.0 0.0 1.0
ca 0.0 0.0 0.0 0.0 0.0 0.0 0.0
cb 2.0 0.0 0.0 2.0 1.0 0.0 0.0
cc 0.0 1.0 0.0 1.0 0.0 1.0 0.0
表 1  模式匹配长度为2时，序列S的模式转移频次和转移概率
图 1  原始的和经符号化处理后的1 min ECG信号片段
图 2  PAF预测模型的总体框架
p n t DIp
1 4.90 5 0.980 0
2 18.90 25 0.750 0
3 51.30 125 0.410 0
4 98.69 625 0.160 0
5 171.25 3 125 0.050 0
6 270.31 15 625 0.020 0
7 398.43 78 125 0.010 0
8 546.88 390 625 0.001 0
9 585.94 1 953 125 0.000 3
10 828.13 9 765 625 <0.000 1
表 2  不同模式匹配长度对应的稠密指数
图 3  CNN-LSTM混合网络模型结构
图 4  基于CNN-LSTM集成预测模型总体结构
图 5  数据流处理示意图
图 6  数据集中各数据类型之间的关系[24]
模型 ACC/% SEN/% SPE/%
1D-CNN 53 68 53
LSTM 66 0 100
1D-CNN-LSTM 61 72 56
本研究方法 91 82 96
表 3  所提方法与基线模型的实验性能对比
文献 特征提取方法 t/min f/Hz ACC/% SEN/% SPE/%
Boon等[4] HRV特征，GA 15 128 79.3 77.4 81.1
10 128 68.8 58.5 81.1
Narin等[5] HRV线性和非线性特征组合，GA 5 128 90.0 92.0 88.0
Sutton等[24] PSPR，模板匹配 1 8 82.1 100.0 73.6
Boon等[26] HRV特征，GAⅢ 5 128 87.7 86.8 88.7
Costin等[27] HRV特征, QRS复合波形态变异性 30 128 89.4 89.3 89.4
本研究方法 PSPR+CNN-LSTM+集成学习 1 8 91.3 82.2 95.8
表 4  模型与相关工作在训练集和测试集随机划分模式下的比较
文献 训练集 测试集 ACC/% SEN/% SPE/%
Thong等[28] ? ? 76.0 68.0 86.0
Zong等[29] ? ? 80.0 ? ?
本研究方法 P06~P25，n11~n50 P01~p05，n01~n10 86.9 82.4 90.0
P01~P05，P11~P25，n01~n10，n21~n50 P06~P10，n11~n20 70.6 60.0 73.1
P01~p10，p16~p25，n01~n20，n31~n50 P11~p15，n31~n30 77.9 76.9 78.2
P01~p15，P21~p25，n01~n30，n41~n50 P16~p20，n31~n40 89.9 89.7 89.9
P01~p20，n01~n40 P21~p25，n41~n50 78.5 67.2 83.1
平均值 80.6 75.7 82.7
表 5  模型在跨患者模式下的实验结果
基学习器 p DIp ACC / %
1 1 0.98 84.09
2 2 0.76 89.33
3 3 0.41 89.52
4 4 0.16 75.64
表 6  基学习器的实验性能对比
f / Hz ACC / % SEN / % SPE / %
128 79.1 69.3 83.7
64 80.2 70.7 84.8
32 78.5 68.6 83.9
16 75.3 69.5 78.1
8 80.6 75.7 82.7
表 7  模型在不同采样频率ECG信号上的实验结果
1 BALL J, CARRINGTON M J, MCMURRAY J J V, et al Atrial fibrillation: profile and burden of an evolving epidemic in the 21st century[J]. International Journal of Cardiology, 2013, 167 (5): 1807- 1824
doi: 10.1016/j.ijcard.2012.12.093
2 黄从新, 张澍, 黄德嘉, 等 心房颤动: 目前的认识和治疗的建议-2018[J]. 中国心脏起搏与心电生理杂志, 2018, 32 (4): 6- 59
HUANG Cong-xin, ZHANG Shu, HUANG De-jia, et al Atrial fibrillation: current understanding and treatment recommendations - 2018[J]. Chinese Journal of Cardiac Pacing and Electrophysiology, 2018, 32 (4): 6- 59
3 黄忠朝, 陈真诚, 赵于前. 基于支持向量机的阵发性房颤自动终止预测研究[C]// 2007中国生物医学工程联合学术年会论文集(上册). 西安: 中国知网. 2007: 180-184.
HUANG Zhong-chao, CEHN Zhen-cheng, ZHAO Yu-qian. Research on automatic termination of paroxysmal atrial fibrillation based on support vector machine [C]// Proceedings of the 2007 China Biomedical Engineering Joint Academic Conference: Volume 1. Xi’an: CNKI, 2007: 180-184.
4 BOON K H, KHALIL-HANI M, MALARVILI M B, et al Paroxysmal atrial fibrillation prediction method with shorter HRV sequences[J]. Computer Methods and Programs in Biomedicine, 2016, 134: 187- 196
doi: 10.1016/j.cmpb.2016.07.016
5 NARIN A, ISLER Y, OZER M, et al Early prediction of paroxysmal atrial fibrillation based on short-term heart rate variability[J]. Physica A: Statistical Mechanics and Its Applications, 2018, 509: 56- 65
doi: 10.1016/j.physa.2018.06.022
6 LAKE D E, MOORMAN J R Accurate estimation of entropy in very short physiological time series: the problem of atrial fibrillation detection in implanted ventricular devices[J]. American Journal of Physiology-Heart and Circulatory Physiology, 2010, 300 (1): 319- 325
7 DEMAZUMDER D, LAKE D E, CHENG A, et al Dynamic analysis of cardiac rhythms for discriminating atrial fibrillation from lethal ventricular arrhythmias[J]. Circulation: Arrhythmia and Electrophysiology, 2013, 6 (3): 555- 561
doi: 10.1161/CIRCEP.113.000034
8 PANUSITTIKORN M, UCHAIPICHAT N, TANTIBUNDHIT C et al. Prediction of paroxysmal atrial fibrillation occurrence with wavelet-based markers [C]// ECTI-CON2010: The 2010 ECTI International Confernce on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology. Chiang Mai: IEEE, 2010: 342-345.
9 POURBABAEE B, LUCAS C. Automatic detection and prediction of paroxysmal atrial fibrillation based on analyzing ecg signal feature classification methods [C]// 2008 Cairo International Biomedical Engineering Conference. Cairo: IEEE, 2008: 1-4.
10 兰天杰, 杨翠微 基于RR间期的阵发性房颤复发预测[J]. 生物医学工程学杂志, 2019, 36 (4): 1- 10
LAN Tian-jie, YANG Cui-wei Prediction of recurrence of paroxysmal atrial fibrillation based on RR interval[J]. Journal of Biomedical Engineering, 2019, 36 (4): 1- 10
11 PARVANEH S, RUBIN J, RAHMAN A, et al Analyzing single-lead short ECG recordings using dense convolutional neural networks and feature-based post-processing to detect atrial fibrillation[J]. Physiological Measurement, 2018, 39 (8): 084003
doi: 10.1088/1361-6579/aad5bd
12 ERDENEBAYAR U, KIM H, PARK J U, et al Automatic prediction of atrial fibrillation based on convolutional neural network using a short-term normal electrocardiogram signal[J]. Journal of Korean Medical Science, 2019, 34 (7): e64
doi: 10.3346/jkms.2019.34.e64
13 MOODY G, GOLDBERGER A, MCCLENNEN S, et al. Predicting the onset of paroxysmal atrial fibrillation: the computers in cardiology challenge 2001 [C]// Computers in Cardiology 2001. Rotterdam: IEEE, 2001: 113-116.
14 GOLDBERGER A L, AMARAL L A, Glass L, et al PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals[J]. Circulation, 2000, 101 (23): 215- 220
15 MOODY G B, MARK R G A new method for detecting atrial fibrillation using R-R intervals[J]. Computers in Cardiology, 1983, 10: 227- 230
16 FAUST O, SHENFIELD A, KAREEM M, et al Automated detection of atrial fibrillation using long short-term memory network with RR interval signals[J]. Computers in Biology and Medicine, 2018, 102: 327- 335
doi: 10.1016/j.compbiomed.2018.07.001
17 KAMALESWARAN R, MAHAJAN R, AKBILGIC O A robust deep convolutional neural network for the classification of abnormal cardiac rhythm using single lead electrocardiograms of variable length[J]. Physiological Measurement, 2018, 39 (3): 035006
doi: 10.1088/1361-6579/aaaa9d
18 RAJPURKAR P, HANNUN A Y, HAGHPANAHI M, et al. Cardiologist-level arrhythmia detection with convolutional neural networks [J/OL]. 2017. [2020-03-13]. https://arxiv.org/pdf/1707.01836.pdf.
19 SCHWAB P, SCEBBA G C, ZHANG J, et al. Beat by beat: classifying cardiac arrhythmias with recurrent neural networks [C]// Computing in Cardiology. Rennes: IEEE, 2017: 1-4.
20 SHASHIKUMAR S P, CLIFFORD G D, SHAH S J, et al. Detection of paroxysmal atrial fibrillation using attention-based bidirectional recurrent neural networks [C]// Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. London: ACM, 2018: 715-723.
21 张异凡, 黄亦翔, 汪开正, 等 用于心律失常识别的LSTM和CNN并行组合模型[J]. 哈尔滨工业大学学报, 2019, 51 (10): 76- 82
ZHANG Yi-fan, HUANG Yi-xiang, WANG Kai-zheng, et al Arrhythmia classification using parallel combination of LSTM and CNN[J]. Journal of Harbin Institute of Technology, 2019, 51 (10): 76- 82
doi: 10.11918/j.issn.0367-6234.201810178
22 MAHAJAN R, KAMALESWARAN R, AKBILGIC O, et al. Effects of varying sampling frequency on the analysis of continuous ECG data streams [C]// VLDB Workshop on Data Management and Analytics for Medicine and Healthcare. Cham: Springer, 2017: 73-87.
23 AKBILGIC O, HOWE J A Symbolic pattern recognition for sequential data[J]. Sequential Analysis, 2017, 36 (4): 528- 540
doi: 10.1080/07474946.2017.1394719
24 SUTTON J R, MAHAJAN R, AKBILGIC O, et al PhysOnline: an open source machinelearning pipeline for real-time analysis of streaming physiological waveform[J]. IEEE Journal of Biomedical and Health Informatics, 2018, 23 (1): 59- 65
25 OH S L, NG E Y K, SAN TAN R, et al Automated diagnosis of arrhythmia using combination of CNN and LSTM techniques with variable length heart beats[J]. Computers in Biology and Medicine, 2018, 102: 278- 287
doi: 10.1016/j.compbiomed.2018.06.002
26 BOON K H, KHALIL-HANI M, MALARVILI M B Paroxysmal atrial fibrillation prediction based on HRV analysis and non-dominated sorting genetic algorithm III[J]. Computer Methods and Programs in Biomedicine, 2018, 153: 171- 184
doi: 10.1016/j.cmpb.2017.10.012
27 COSTIN H, ROTARIU C, P?S?RIC? A. Atrial fibrillation onset prediction using variability of ECG signals [C]// 2013 8th International Symposium on Advanced Topics in Electrical Engineering. Bucharest: IEEE, 2013: 1-4.
28 THONG T, MCNAMES J, ABOY M, et al. Paroxysmal atrial fibrillation prediction using isolated premature atrial events and paroxysmal atrial tachycardia [C]// Proceedings of the 25th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Cancun: IEEE, 2003, 1: 163-166.
[1] 许佳辉,王敬昌,陈岭,吴勇. 基于图神经网络的地表水水质预测模型[J]. 浙江大学学报(工学版), 2021, 55(4): 601-607.
[2] 于勇,薛静远,戴晟,鲍强伟,赵罡. 机加零件质量预测与工艺参数优化方法[J]. 浙江大学学报(工学版), 2021, 55(3): 441-447.
[3] 毛奕喆,龚国芳,周星海,王飞. 基于马尔可夫过程和深度神经网络的TBM围岩识别[J]. 浙江大学学报(工学版), 2021, 55(3): 448-454.
[4] 黄发明,曹中山,姚池,姜清辉,陈佳武. 基于决策树和有效降雨强度的滑坡危险性预警[J]. 浙江大学学报(工学版), 2021, 55(3): 472-482.
[5] 苗发盛,吴益平,李麟玮,廖康,薛阳. 基于Boosting-决策树C5.0的岩体结构面粗糙度预测[J]. 浙江大学学报(工学版), 2021, 55(3): 483-490.
[6] 刘芳,汪震,刘睿迪,王锴. 基于组合损失函数的BP神经网络风力发电短期预测方法[J]. 浙江大学学报(工学版), 2021, 55(3): 594-600.
[7] 段有康,陈小刚,桂剑,马斌,李顺芬,宋志棠. 基于相位划分的下肢连续运动预测[J]. 浙江大学学报(工学版), 2021, 55(1): 89-95.
[8] 陈纬奇,王敬昌,陈岭,杨勇勤,吴勇. 基于深度神经网络的多因素感知终端换机预测模型[J]. 浙江大学学报(工学版), 2021, 55(1): 109-115.
[9] 陈巧红,陈翊,李文书,贾宇波. 多尺度SE-Xception服装图像分类[J]. 浙江大学学报(工学版), 2020, 54(9): 1727-1735.
[10] 李文书,邹涛涛,王洪雁,黄海. 基于双尺度长短期记忆网络的交通事故量预测模型[J]. 浙江大学学报(工学版), 2020, 54(8): 1613-1619.
[11] 张彦楠,黄小红,马严,丛群. 基于深度学习的录音文本分类方法[J]. 浙江大学学报(工学版), 2020, 54(7): 1264-1271.
[12] 江冬冬,李道飞,俞小莉. 基于驾驶员需求转矩预测的模型预测控制能量管理[J]. 浙江大学学报(工学版), 2020, 54(7): 1325-1334.
[13] 王晨霖,杨洁,居文军,顾复,陈芨熙,纪杨建. 基于智能家电的短期电力负荷预测与削峰填谷优化[J]. 浙江大学学报(工学版), 2020, 54(7): 1418-1424.
[14] 闫旭,范晓亮,郑传潘,臧彧,王程,程明,陈龙彪. 基于图卷积神经网络的城市交通态势预测算法[J]. 浙江大学学报(工学版), 2020, 54(6): 1147-1155.
[15] 刘创,梁军. 基于注意力机制的车辆运动轨迹预测[J]. 浙江大学学报(工学版), 2020, 54(6): 1156-1163.