CLECG: A Novel Contrastive Learning Framework for Electrocardiogram Arrhythmia Classification
Problem
ECG diagnosis algorithms need large labeled datasets, but privacy concerns and annotation costs cause a shortage of labeled ECG data.
Why it matters
Limited training data restricts automated heart arrhythmia diagnosis, affecting cardiac care quality.
What the AI did
Proposed CLECG, an instance-level contrastive learning scheme that pre-trains on unlabeled ECG signals by pulling augmented views of the same signal together and pushing different signals apart, requiring no labels.
Results
Stated to outperform other self-supervised methods and supervised transfer learning; no concrete numbers given
Impact
Evidence supports a labeling-free pre-training method that improves ECG classification over compared baselines.
Record
| Venue | IEEE Signal Processing Letters |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Hui Chen, Guijin Wang, Guodong Zhang, Ping Zhang, Huazhong Yang |
| Citations | 35 (OpenAlex, at import time) |
| References | 45 |
| Model or method | CLECG (contrastive learning framework) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1109/lsp.2021.3114119 |
| Open access | closed |
| Keywords | ECG, contrastive learning, self-supervised, arrhythmia, unlabeled data |
Importance rating
real but narrower advance
Label shortage limits automated arrhythmia diagnosis; a labeling-free pre-training method is a meaningful methodological advance for cardiac care, but no scale is stated.