Objective:

To develop an ECG arrhythmia classification framework that achieves strong performance under clean and noisy conditions, real-time inference with low resource consumption, and privacy-preserving training capability.

Approach:

• Model Development: TransECG-Net, a hybrid CNN–Transformer network, was developed for AAMI-aligned five-class heartbeat classification.
• Feature Extraction: The CNN branch extracts local morphology features, while the Transformer branch models global temporal dependencies.
• Data Handling: Public ECG recordings were segmented into fixed-length heartbeat windows and split into training, validation, and testing subsets.
• Performance Assessment: Model performance was evaluated using accuracy, precision, recall, specificity, and F1-score.

Key Findings:

• TransECG-Net achieved 99.52% accuracy on testing samples.
• Class-wise F1-scores were 99.90% for N, 99.43% for L, 99.56% for R, 99.61% for A, and 99.12% for V.
• TransECG-Net outperformed DeepECG-Net (98.30%) and Hybrid CNN-BLSTM (94.20%).
• The model maintained 35 ms latency and a 28 MB memory footprint.

Interpretation:

TransECG-Net supports accurate, physiology-guided, noise-tolerant, and edge-deployable ECG arrhythmia screening for wearable and clinical monitoring.

Limitations:

• The study does not address the performance of the model in real-world clinical settings.
• Potential challenges in federated learning with heterogeneous, non-IID data distributions are not fully explored.

Conclusion:

TransECG-Net provides a scalable foundation for AI-assisted ECG screening systems that can support continuous monitoring and enhance early abnormality detection.