中国邮电高校学报(英文) ›› 2022, Vol. 29 ›› Issue (6): 83-96.doi: 10.19682/j.cnki.1005-8885.2022.1008

• Others • 上一篇    

Research on ECG classification based on transfer learning

Jiang Fan, Chen Jiajun, Gao Youjun, Sun Changyin   

  1. 1. Shaanxi Key Laboratory of Information Communication Network and Security, Xi'an University of Posts and Telecommunications, Xi'an 710121, China 2. China Mobile System Integration Co. , Ltd. , Beijing 100052, China 3. China Mobile Xiong'an Information Communication Technology Co. , Ltd. , Beijing 100052, China
  • 收稿日期:2021-04-23 修回日期:2021-12-13 出版日期:2022-12-30 发布日期:2022-12-30
  • 通讯作者: Chen Jiajun E-mail:chen_jj123@126.com
  • 基金资助:

    This work was supported by the National Natural Science Foundation of China ( 62071377 ), and the Key Project of

    Natural Science Foundation of Shaanxi Province (2019ZDLGY07-06, 2021JM-465).

Research on ECG classification based on transfer learning

Jiang Fan, Chen Jiajun, Gao Youjun, Sun Changyin   

  1. 1. Shaanxi Key Laboratory of Information Communication Network and Security, Xi'an University of Posts and Telecommunications, Xi'an 710121, China 2. China Mobile System Integration Co. , Ltd. , Beijing 100052, China 3. China Mobile Xiong'an Information Communication Technology Co. , Ltd. , Beijing 100052, China
  • Received:2021-04-23 Revised:2021-12-13 Online:2022-12-30 Published:2022-12-30
  • Contact: Chen Jiajun E-mail:chen_jj123@126.com
  • Supported by:

    This work was supported by the National Natural Science Foundation of China ( 62071377 ), and the Key Project of

    Natural Science Foundation of Shaanxi Province (2019ZDLGY07-06, 2021JM-465).

摘要:

Concerning current deep learning-based electrocardiograph ( ECG) classification methods, there exists domain discrepancy between the data distributions of the training set and the test set in the inter-patient paradigm. To reduce the negative effect of domain discrepancy on the classification accuracy of ECG signals, this paper incorporates transfer learning into the ECG classification, which aims at applying the knowledge learned from the raining set to the test set. Specifically, this paper first develops a deep domain adaptation network ( DAN) for ECG classification based on the convolutional neural network ( CNN). Then, the network is pre-trained with training set data obtained from the famous Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) ECG arrhythmia database. On this basis, by minimizing the multi-kernel maximum mean discrepancy (MK-MMD) between the data distributions of the training set and the test set, the pre-trained network is adjusted to learn transferable feature representations. Finally, with the low-density separation of unlabeled target data, the feature representations are more transferable. The extensive experimental results show that the proposed domain adaptation method has reached a 7. 58% improvement in overall classification accuracy on the test set, and achieves competitive performance with other state-of-the-arts.

关键词:

electrocardiograph (ECG) classification, transfer learning, domain adaptation, convolutional neural network (CNN)

Abstract: Concerning current deep learning-based electrocardiograph ( ECG) classification methods, there exists domain discrepancy between the data distributions of the training set and the test set in the inter-patient paradigm. To reduce the negative effect of domain discrepancy on the classification accuracy of ECG signals, this paper incorporates transfer learning into the ECG classification, which aims at applying the knowledge learned from the raining set to the test set. Specifically, this paper first develops a deep domain adaptation network ( DAN) for ECG classification based on the convolutional neural network ( CNN). Then, the network is pre-trained with training set data obtained from the famous Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) ECG arrhythmia database. On this basis, by minimizing the multi-kernel maximum mean discrepancy (MK-MMD) between the data distributions of the training set and the test set, the pre-trained network is adjusted to learn transferable feature representations. Finally, with the low-density separation of unlabeled target data, the feature representations are more transferable. The extensive experimental results show that the proposed domain adaptation method has reached a 7. 58% improvement in overall classification accuracy on the test set, and achieves competitive performance with other state-of-the-arts.

Key words: electrocardiograph (ECG) classification, transfer learning, domain adaptation, convolutional neural network (CNN)

中图分类号: