An adaptive multi-scale dilated convolutional neural network for arrhythmia detection
The rising global incidence of arrhythmia calls for reliable detection methods. Most existing approaches suffer from poor generalization and insufficient feature extraction for long-term electrocardiogram (ECG) signals. This work proposes an adaptive multi-scale dilated convolutional network (AMSDNet). It uses multi-scale dilated convolutions, gated adaptive weight allocation, and spatial attention fusion to capture global ECG features. Validated on Massachusetts Institute of Technology-Beth Israel Hospital Arrhythmia Database (MIT-BIH), the China Physiological Signal Challenge (CPSC) 2018, and the Computing in Cardiology Challenge (CINC) 2017 datasets, our method outperforms state-of-the-art models with average accuracy of 99.7%, 91.8% and 83.7%, respectively.
Authors
- Yunfan Chen (ORCID: https://orcid.org/0000-0003-4808-6352)
- 吴植强 WU Zhiqiang
- Yuting Li
- Panpan Hu
- Yexin Wan
Institutions
- Hubei University of Technology (CN)
Publication Details
- Journal
- Computer Methods in Biomechanics & Biomedical Engineering
- Published
- 2026-11-12
- DOI
- https://doi.org/10.1080/10255842.2026.2746753
- Primary Topic
- ECG Monitoring and Analysis
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China