Physiology-guided beat-level arrhythmia classification from ECG using a CNN-transformer hybrid neural network - Report - MDSpire

Physiology-guided beat-level arrhythmia classification from ECG using a CNN-transformer hybrid neural network

  • By

  • Guangfeng Li

  • Zhidong Zhang

  • Gang Qiao

  • Xiaosan Chen

  • Gangqiang Zhou

  • Kavimbi Chipusu

  • July 8, 2026

  • 0 min

Share

Clinical Report: Physiology-Informed Classification of Arrhythmias from ECG Data

Overview

TransECG-Net, a hybrid CNN-Transformer network, achieved 99.52% accuracy in classifying ECG arrhythmias across five classes.

Background

Accurate classification of arrhythmias via ECG is vital for effective screening and monitoring, particularly as cardiovascular diseases remain a leading cause of morbidity and mortality. Traditional manual interpretation of ECGs is time-consuming and requires significant expertise, which limits scalability in high-throughput settings.

Data Highlights

MetricValue
Accuracy99.52%
F1-score (N)99.90%
F1-score (L)99.43%
F1-score (R)99.56%
F1-score (A)99.61%
F1-score (V)99.12%
Macro-averaged F1-score99.52%
Latency35 ms
Memory Footprint28 MB

Key Findings

  • TransECG-Net achieved 99.52% accuracy in classifying ECG arrhythmias.
  • Class-wise F1-scores ranged from 99.12% to 99.90% across five classes.
  • The model outperformed DeepECG-Net (98.30%) and Hybrid CNN-BLSTM (94.20%).
  • It maintained a low latency of 35 ms and a memory footprint of 28 MB.
  • The hybrid architecture combines CNN and Transformer features for improved classification.

Clinical Implications

The TransECG-Net model provides a solution for ECG arrhythmia classification.

Conclusion

TransECG-Net offers a tool for ECG arrhythmia classification.

Related Resources & Content

  1. HRS scientific statement on artificial intelligence integration framework into clinical electrophysiology workflows - PubMed, 2026 -- Guidance and consensus on AI in electrophysiology
  2. npj Digital Medicine — Interpretable arrhythmia detection in ECG scans using deep learning ensembles: a genetic programming approach, 2025 -- Deep learning for arrhythmia detection
  3. npj Digital Medicine — Bridging clinical knowledge and AI: an interpretable transformer framework for ECG diagnosis, 2025 -- AI in ECG diagnosis
  4. npj Digital Medicine — Deriving novel atrial fibrillation phenotypes using a tree-based artificial intelligence-enhanced electrocardiography approach, 2025 -- Atrial fibrillation phenotyping
  5. 26-A-12312-ACC WEARABLE PHOTOPLETHYSMOGRAPHY AND AI FOR ATRIAL FIBRILLATION SCREENING: PROSPECTIVE AND RANDOMIZED EVIDENCE | JACC, 2026 -- Evidence on AI-enabled rhythm detection
  6. npj Digital Medicine — xGNN4MI: Enhancing Interpretability of Graph Neural Networks in 12-Lead ECG for Cardiovascular Disease Assessment
  7. HRS scientific statement on artificial intelligence integration framework into clinical electrophysiology workflows - PubMed
  8. 26-A-12312-ACC WEARABLE PHOTOPLETHYSMOGRAPHY AND AI FOR ATRIAL FIBRILLATION SCREENING: PROSPECTIVE AND RANDOMIZED EVIDENCE | JACC
  9. ANSI/AAMI EC57:2012/(R)2020 (PDF) | AAMI

Original Source(s)

Physiology-guided beat-level arrhythmia classification from ECG using a CNN-transformer hybrid neural network

  • by Guangfeng Li, Zhidong Zhang, Gang Qiao, Xiaosan Chen, Gangqiang Zhou, Kavimbi Chipusu

  • July 8, 2026

Related Content

Self-supervised representation learning reveals explainable physiological structure in high-dimensional magnetocardiography

FDA Approves Dual-Energy Catheter Ablation Platform

The platform integrates with the CARTO ecosystem and enables physicians to use either pulsed field or radiofrequency energy based on patient anatomy and case complexity.

Research hotspots and trend of the heart-brain axis by MRI: a bibliometric analysis