许越凡等：基于一维卷积特征与手工特征融合的集成超限学习机心跳分类方法 1231· 表3混淆矩阵 Table 3 Confusion matrix Predict labels L R A IF x j f E JeQ N 64996 9 3 58 126 1 0 82843240001 L 6 4829 0 0 7 0 0 0 0 00 01000 R 8 0 4336 6 0 000 00000 A 169 20 1321 0 0 6 10000 62 0 0 10000 60000 0000 0 0000 True labels 0000 0 0000 12 0 0100 4500000 E 0 0 50000 1 0 0 03700 e 7 0 0 0 0 0 00010 0 0 0 0 0 20001 表4提出的方法与其他方法的比较结果 [3] Mondejar-Guerra V,Novo J,Rouco J,et al.Heartbeat classification fusing temporal and morphological information of Table 4 Comparison results of the proposed approach with other approaches ECGs via ensemble of classifiers.Biomed Signal Process Control, Accuracy/ 2019,47:41 Reference Features Classifier 9% [4] Chazal P d,O'Dwyer M,Reilly R B.Automatic classification of Manual features only DWT,RR+ELM (Single) 98.28 heartbeats using ECG morphology and heartbeat interval features. Deep feature only 1D CNN 98.50 IEEE Trans Biomed Eng,2004,51(7):1196 Feature fusion DWT.RR,ID Convolution+ 98.81 [5] Tuncer T,Dogan S,Plawiak P,et al.Automated arrhythmia (Without ensemble) ELM (Single) Yel ICA.Wavelet,RR+SVM detection using novel hexadecimal local pattem and multilevel (One-against-one) 98.72 wavelet transform with ECG signals.Know/Based Syst,2019, Our proposed approach DWT,RR,ID Convolution ELM 99.02 186:104923 (Bagging ensemble) [6]Sahoo S,Kanungo B.Behera S,et al.Multiresolution wavelet 并得到更好的分类结果.此外，基于Bagging策略 transform based feature extraction and ECG classification to detect 的ELM集成方法可进一步提升心跳分类性能.未 cardiac abnormalities.Measurement,2017,108:55 [7]Afkhami R G,Azarnia G,Tinati M A.Cardiac arrhythmia 来工作将基于其他公开的心律失常数据集，对所 classification using statistical and mixture modeling features of 提出的方法加以改进.同时，将所提方法应用到临 ECG signals.Pattern Recognit Lett,2016,70:45 床实践中 [8]Marinho L B,Nascimento N MM,Souza J W M,et al.A novel electrocardiogram feature extraction approach for cardiac 参考文献 arrhythmia classification.Future Gener Comput Syst,2019,97: [1]Thaler M S.The Only EKG Book You'll Ever Need.9th Ed. 564 Philadelphia,PA:Lippincott Williams Wilkins,2017[2] [9]Ye C,Vijaya Kumar B V K,Coimbra M T.Heartbeat Gawlowska J,Wranicz JK,Norman J."Jeff"Holter (1914-1983). classification using morphological and dynamic features of ECG Cardiol J,2009,16(4):386 signals.IEEE Trans Biomed Eng,2012,59(10):2930 [2]Chazal P d,Reilly R B.A patient-adapting heartbeat classifier [10]Castillo O,Melin P,Ramirez E,et al.Hybrid intelligent system for using ECG morphology and heartbeat interval features.IEEE cardiac arrhythmia classification with Fuzzy K-Nearest Neighbors Trans Biomed Eng,2006,53(12):2535 and neural networks combined with a fuzzy system.Expert Syst并得到更好的分类结果. 此外，基于 Bagging 策略 的 ELM 集成方法可进一步提升心跳分类性能. 未 来工作将基于其他公开的心律失常数据集，对所 提出的方法加以改进. 同时，将所提方法应用到临 床实践中. 参    考    文    献 Thaler  M  S.  The  Only  EKG  Book  You ’ll  Ever  Need.  9th  Ed. Philadelphia,  PA:  Lippincott  Williams  &  Wilkins,  2017[2] Gawlowska J, Wranicz J K, Norman J. “Jeff” Holter (1914—1983). Cardiol J, 2009, 16(4): 386 [1] Chazal  P  d,  Reilly  R  B.  A  patient-adapting  heartbeat  classifier using  ECG  morphology  and  heartbeat  interval  features. IEEE Trans Biomed Eng, 2006, 53（12）: 2535 [2] Mondéjar-Guerra  V,  Novo  J,  Rouco  J,  et  al.  Heartbeat classification  fusing  temporal  and  morphological  information  of ECGs via ensemble of classifiers. Biomed Signal Process Control, 2019, 47: 41 [3] Chazal  P  d,  O'Dwyer  M,  Reilly  R  B.  Automatic  classification  of heartbeats using ECG morphology and heartbeat interval features. IEEE Trans Biomed Eng, 2004, 51（7）: 1196 [4] Tuncer  T,  Dogan  S,  Pławiak  P,  et  al.  Automated  arrhythmia detection  using  novel  hexadecimal  local  pattern  and  multilevel wavelet  transform  with  ECG  signals. Knowl Based Syst,  2019, 186: 104923 [5] Sahoo  S,  Kanungo  B,  Behera  S,  et  al.  Multiresolution  wavelet transform based feature extraction and ECG classification to detect cardiac abnormalities. Measurement, 2017, 108: 55 [6] Afkhami  R  G,  Azarnia  G,  Tinati  M  A.  Cardiac  arrhythmia classification  using  statistical  and  mixture  modeling  features  of ECG signals. Pattern Recognit Lett, 2016, 70: 45 [7] Marinho L B, Nascimento N M M, Souza J W M, et al. A novel electrocardiogram  feature  extraction  approach  for  cardiac arrhythmia  classification. Future Gener Comput Syst,  2019,  97: 564 [8] Ye  C,  Vijaya  Kumar  B  V  K,  Coimbra  M  T.  Heartbeat classification  using  morphological  and  dynamic  features  of  ECG signals. IEEE Trans Biomed Eng, 2012, 59（10）: 2930 [9] Castillo O, Melin P, Ramírez E, et al. Hybrid intelligent system for cardiac arrhythmia classification with Fuzzy K-Nearest Neighbors and  neural  networks  combined  with  a  fuzzy  system. Expert Syst [10] 表 3 混淆矩阵 Table 3 Confusion matrix Predict labels N L R A V / a ! F x j f E J e Q True labels N 64996 9 3 58 126 1 0 8 28 4 32 4 0 0 0 1 L 6 4829 0 0 7 0 0 0 0 0 0 0 1 0 0 0 R 8 0 4336 6 2 0 1 0 0 0 0 0 0 0 0 0 A 169 1 20 1321 9 0 0 1 0 0 6 1 0 0 0 0 V 62 4 1 2 4183 0 1 5 18 0 0 1 0 0 0 0 / 5 0 0 0 1 4203 0 0 0 0 1 6 0 0 0 0 a 11 2 0 4 7 0 49 2 0 0 0 0 0 0 0 0 ! 8 0 0 0 5 0 1 221 0 1 0 0 0 0 0 0 F 54 0 1 0 27 0 0 0 319 0 0 0 0 0 0 0 x 4 0 0 1 4 0 1 2 0 84 1 0 0 0 0 0 j 12 0 1 0 0 0 0 0 0 0 101 0 0 1 0 0 f 16 0 0 0 1 24 0 0 0 0 0 450 0 0 0 0 E 1 0 0 0 2 0 0 0 0 0 0 0 50 0 0 0 J 1 0 2 0 0 0 0 0 0 0 1 0 0 37 0 0 e 7 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 Q 10 0 1 0 3 0 0 0 0 0 0 2 0 0 0 1 表 4    提出的方法与其他方法的比较结果 Table 4    Comparison  results  of  the  proposed  approach  with  other approaches Reference Features + Classifier Accuracy/ % Manual features only DWT, RR + ELM (Single) 98.28 Deep feature only 1D CNN 98.50 Feature fusion (Without ensemble) DWT, RR, 1D Convolution + ELM (Single) 98.81 Ye[9] ICA, Wavelet, RR + SVM (One-against-one) 98.72 Our proposed approach DWT, RR, 1D Convolution + ELM (Bagging ensemble) 99.02 许越凡等： 基于一维卷积特征与手工特征融合的集成超限学习机心跳分类方法 · 1231 ·