Deep Discriminative Domain Generalization with Adversarial Feature Learning for Classifying ECG Signals

Zuogang Shang1, Zhibin Zhao2, Hui Fang3, Samuel Relton4, Darcy Murphy5, Zoe Hancox4, Ruqiang Yan2, David Wong5
1Xi’an Jiaotong University, 2the Department of Mechanical Engineering, Xi’an Jiaotong University, 3the Department of Computer Science, Loughborough University, 4Leeds Institute of Health Sciences, University of Leeds, 5the Department of Computer Science and the Center for Health Informatics, University of Mancheste


Introduction: The goal of the 2021 PhysioNet/CinC challenge is diagnosing cardiac abnormalities from electrocardiograms (ECGs) and evaluating the diagnostic potential of reduced-lead ECGs. We describe the whole model created by the team ``AI\_Healthcare" for this goal.

  Methods: ECGs were downsampled to 300 Hz and filtered by wavelet. Then ECGs we randomly clipped or zero-padded to 4,096 samples. To have a better representative learning ability, a modified ResNet with larger kernel sizes was used. The multi-source adversarial feature learning was used to learn domain-invariant and discriminative representations with a special gradient reversal layer (GRL). %Different thresholds were used and ECGs were processed multiple times and the average results were recorded. 
  The performance with and without the domain generation methods is compared.

  Results: We achieved a challenge score of 0.66, 0.64, 0.65, 0.65, 0.62 on the validation data. We ranked 8th, 7th, 6th, 6th, and 12th for 12-lead, 6-lead, 4-lead, 3-lead, and 2-lead ECGs, respectively. Testing showed that domain generation improved metric scores on the unseen domain.

  Conclusion: Generalized representations perform well for “unseen” data. It is a general method for other models to improve generalization performance by learning a domain-invariant feature representation.