J Shanghai Jiaotong Univ Sci

Journal of Shanghai Jiaotong University(Science) >

2022 , Vol. 27 >Issue 1: 55 - 69

DOI: https://doi.org/10.1007/s12204-021-2371-8

Robotics & AI in Interdisciplinary Medicine and Engineering

Multi-Lead ECG Classification via an Information-Based Attention Convolutional Neural Network

Expand
  • (1. School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China; 2. Shukun (Beijing) Network Technology Co., Ltd., Beijing 100190, China; 3. Institute of Medical Robotics, Shanghai Jiao Tong University, Shanghai 200240, China)

Received date: 2020-03-30

  Online published: 2022-01-14

Fold

Abstract

A novel structure based on channel-wise attention mechanism is presented in this paper. With the proposed structure embedded, an efficient classification model that accepts multi-lead electrocardiogram (ECG) as input is constructed. One-dimensional convolutional neural networks (CNNs) have proven to be effective in pervasive classification tasks, enabling the automatic extraction of features while classifying targets. We implement the residual connection and design a structure which can learn the weights from the information contained in different channels in the input feature map during the training process. An indicator named mean square deviation is introduced to monitor the performance of a particular model segment in the classification task on the two out of five ECG classes. The data in the MIT-BIH arrhythmia database is used and a series of control experiments is conducted. Utilizing both leads of the ECG signals as input to the neural network classifier can achieve better classification results than those from using single channel inputs in different application scenarios. Models embedded with the channel-wise attention structure always achieve better scores on sensitivity and precision than the plain Resnet models. The proposed model exceeds most of the state-of-the-art models in ventricular ectopic beats (VEB) classification performance and achieves competitive scores for supraventricular ectopic beats (SVEB). Adopting more lead ECG signals as input can increase the dimensions of the input feature maps, helping to improve both the performance and generalization of the network model. Due to its end-to-end characteristics, and the extensible intrinsic for multi-lead heart diseases diagnosing, the proposed model can be used for the realtime ECG tracking of ECG waveforms for Holter or wearable devices.

Key words： attention model, convolutional neural network (CNN), ECG classification, arrhythmia

Cite this article

TUNG Hao (董昊), ZHENG Chao (郑超), MAO Xinsheng(毛新生), QIAN Dahong (钱大宏) . Multi-Lead ECG Classification via an Information-Based Attention Convolutional Neural Network[J]. Journal of Shanghai Jiaotong University(Science), 2022 , 27(1) : 55 -69 . DOI: 10.1007/s12204-021-2371-8

References

[1] JAYARAMAN S, SANGAREDDI V, PERIYASAMYR, et al. Modified limb lead ECG system effects on electrocardiographicwave amplitudes and frontal planeaxis in sinus rhythm subjects [J]. The Anatolian Journalof Cardiology, 2017, 17(1): 46-54. [2] MARK R, MOODY G. MIT-BIH arrhythmia databasedirectory [M]. 2nd ed. Cambridge, MA: MIT Press,1988. [3] OSOWSKI S, HOAI L T, MARKIEWICZ T. Supportvector machine-based expert system for reliable heartbeatrecognition [J]. IEEE Transactions on BiomedicalEngineering, 2004, 51(4): 582-589. [4] COAST D A, STERN R M, CANO G G, et al. Anapproach to cardiac arrhythmia analysis using hiddenMarkov models [J]. IEEE Transactions on BiomedicalEngineering, 1990, 37(9): 826-836. [5] AFONSO V X, TOMPKINS W J, NGUYEN T Q, etal. ECG beat detection using filter banks [J]. IEEETransactions on Biomedical Engineering, 1999, 46(2):192-202. [6] INAN O T, GIOVANGRANDI L, KOVACS G T A.Robust neural-network-based classification of prematureventricular contractions using wavelet transformand timing interval features [J]. IEEE Transactions onBiomedical Engineering, 2006, 53(12): 2507-2515. [7] DE CHAZAL P, O’DWYER M, REILLY R B. Automaticclassification of heartbeats using ECG morphologyand heartbeat interval features [J]. IEEE Transactionson Biomedical Engineering, 2004, 51(7): 1196-1206. [8] DAS M K, ARI S. ECG beats classification using mixtureof features [J]. International Scholarly ResearchNotices, 2014, 2014: 178436. [9] SAHOO S, KANUNGO B, BEHERA S, et al. Multiresolutionwavelet transform based feature extraction andECG classification to detect cardiac abnormalities [J].Measurement, 2017, 108: 55-66. [10] KIRANYAZ S, INCE T, GABBOUJ M. Real-timepatient-specific ECG classification by 1-D convolutionalneural networks [J]. IEEE Transactions onBiomedical Engineering, 2016, 63(3): 664-675. [11] XU S S, MAKMW, CHEUNG C C. Towards end-toendECG classification with raw signal extraction anddeep neural networks [J]. IEEE Journal of Biomedicaland Health Informatics, 2019, 23(4): 1574-1584. [12] ACHARYA U R, FUJITA H, LIH O S, et al. Automateddetection of arrhythmias using different intervalsof tachycardia ECG segments with convolutionalneural network [J]. Information Sciences, 2017, 405:81-90. [13] TAN J H, HAGIWARA Y, PANG W, et al. Applicationof stacked convolutional and long short-termmemory network for accurate identification of CADECG signals [J]. Computers in Biology and Medicine,2018, 94: 19-26. [14] KUMAR M, PACHORI R B, ACHARYA U R. Characterizationof coronary artery disease using flexibleanalytic wavelet transform applied on ECG signals [J].Biomedical Signal Processing and Control, 2017, 31:301-308. [15] ACHARYA U R, SUDARSHAN V K, KOH J E W, etal. Application of higher-order spectra for the characterizationof Coronary artery disease using electrocardiogramsignals [J]. Biomedical Signal Processing andControl, 2017, 31: 31-43. [16] ACHARYA U R, FUJITA H, LIH O S, et al. Automateddetection of coronary artery disease using differentdurations of ECG segments with convolutionalneural network [J]. Knowledge-Based Systems, 2017,132: 62-71. [17] MIRVIS D M, BERSON A S, GOLDBERGER A L,et al. Instrumentation and practice standards for electrocardiographicmonitoring in special care units. Areport for health professionals by a Task Force of theCouncil on Clinical Cardiology, American Heart Association[J]. Circulation, 1989, 79(2): 464-471. [18] SU J, DAI J, GUAN Z, et al. A four-lead real time arrhythmiaanalysis algorithm [C]//2017 Computing inCardiology Conference (CinC ). Rennes: IEEE, 2017:1-4. [19] YAN Y, QIN X B,WU Y G, et al. A restricted Boltzmannmachine based two-lead electrocardiography classification [C]//2015 IEEE 12th International Conferenceon Wearable and Implantable Body Sensor Networks(BSN). Cambridge, MA: IEEE, 2015: 1-9. [20] CHAZAL P. Different techniques used to improve theperformance of a classifier of the twelve-lead electrocardiogram[C]//Computers in Cardiology 2001. Rotterdam:IEEE, 2001: 525-528. [21] LA FW, TSAI P Y. Deep learning for detection of fetalECG from multi-channel abdominal leads [C]//2018Asia-Pacific Signal and Information Processing AssociationAnnual Summit and Conference (APSIPAASC). Honolulu, HI: IEEE, 2018: 1397-1401. [22] LUZ E J D S, SCHWARTZWR, C′AMARA-CH′AVEZG, et al. ECG-based heartbeat classification for arrhythmiadetection: A survey [J]. Computer Methodsand Programs in Biomedicine, 2016, 127: 144-164. [23] YE C, KUMAR B V K V, COIMBRA M T. An automaticsubject-adaptable heartbeat classifier based onmultiview learning [J]. IEEE Journal of Biomedicaland Health Informatics, 2016, 20(6): 1485-1492. [24] INCE T, KIRANYAZ S, GABBOUJ M. A generic androbust system for automated patient-specific classificationof ECG signals [J]. IEEE Transactions on BiomedicalEngineering, 2009, 56(5): 1415-1426. [25] JIANG W, KONG S G. Block-based neural networksfor personalized ECG signal classification [J]. IEEETransactions on Neural Networks, 2007, 18(6): 1750-1761. [26] HU Y H, PALREDDY S, TOMPKINSWJ. A patientadaptableECG beat classifier using a mixture of expertsapproach [J]. IEEE Transactions on BiomedicalEngineering, 1997, 44(9): 891-900. [27] AAMI. Testing and reporting performance results ofventricular arrhythmia detection algorithms [S]. Arlington,VA: Association Advancement Medical Instrumentation,1987. [28] TEIJEIRO T, F′ELIX P, PRESEDO J, et al. Heartbeatclassification using abstract features from the abductiveinterpretation of the ECG [J]. IEEE Journal ofBiomedical and Health Informatics, 2018, 22(2): 409-420. [29] MOODY G B, MARK R G. The impact of the mitbiharrhythmia database [J]. IEEE Engineering inMedicine and Biology Magazine, 2001, 20(3): 45-50. [30] GOLDBERGER A L, AMARAL L A, GLASS L, etal. PhysioBank, PhysioToolkit, and PhysioNet: Componentsof a new research resource for complex physiologicsignals [J]. Circulation, 2000, 101(23): E215-E220. [31] SABHERWAL P, AGRAWAL M, SINGH L. Automaticdetection of the R peaks in single-lead ECG signal[J]. Circuits, Systems, and Signal Processing, 2017,36(11): 4637-4652. [32] PAN J, TOMPKINS W J. A real-time QRS detectionalgorithm [J]. IEEE Transactions on Biomedical Engineering,1985, 32(3): 230-236. [33] LI C, ZHENG C, TAI C. Detection of ECG characteristicpoints using wavelet transforms [J]. IEEE Transactionson Biomedical Engineering, 1995, 42(1): 21-28. [34] YE C, VIJAYA KUMAR B V K, COIMBRA M T.Heartbeat classification using morphological and dynamicfeatures of ECG signals [J]. IEEE Transactionson Biomedical Engineering, 2012, 59(10): 2930-2941. [35] WEI X C. Analyze deep learning: CNN theory andvisual practice [M]. Beijing: Publishing House of ElectronicsIndustry, 2018: 137-138 (in Chinese). [36] CHAWLA N V, BOWYER K W, HALL L O, et al.SMOTE: Synthetic minority over-sampling technique[J]. Journal of Artificial Intelligence Research, 2002,16: 321-357. [37] BATISTA G E A P A, BAZZAN A L C,MONARDMC. Balancing training data for automated annotationof keywords: A case study [C]//II Brazilian Workshopon Bioinformatics. Maca′e RJ: DBLP, 2003: 10-18. [38] HE K, ZHANG X, REN S, et al. Identity mappings indeep residual networks [M]//Computer vision – ECCV2016. Cham: Springer, 2016: 630-645. [39] HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]//2018 IEEE/CVF Conference on ComputerVision and Pattern Recognition. Salt Lake City, UT:IEEE, 2018: 7132-7141. [40] WANG J. Proposed new requirements for testing andreporting performance results of arrhythmia detectionalgorithms [J]. Journal of Electrocardiology, 2014,47(6): 909. [41] SCHWARTZ-ZIV R, TISHBY N. Opening the blackbox of deep neural networks via information [EB/OL].(2017-04-29). https://arxiv.org/pdf/1703.00810. [42] TISHBY N, ZASLAVSKY N. Deep learning and theinformation bottleneck principle [J]. 2015 IEEE InformationTheory Workshop (ITW). Jerusalem: IEEE,2015: 1-5.
Options
Outlines

/

〈 〉