A Dual-attention Deep Learning Approach for Real-time ECG Arrhythmia Detection under Multiclass Imbalance

Read full text Summarise with VIRAT
Authors: Soumia Zertal, Asma Saighi, Jorge García–Gutiérrez
Cited by: 0
References: 48

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Cardiovascular diseases continue to be a health issue, highlighting the importance of real-time and accurate arrhythmia detection from ECG signals.Wearable and embedded systems require models that capture morphological and long-range temporal features while maintaining low computational overhead.Existing approaches suffer from severe class imbalance and a limited class coverage, which reduces clinical applicability.This paper provides an approach called a Dual-Attention enhanced Network for ElectroCardioGram Classification (DAN-ECGC), to detect 23 arrhythmias types in real time.The model combines a squeeze-and-excitation convolutional encoder with a bidirectional gated recurrent unit and multi-head attention to extract both local and global electrocardiogram patterns.A weighted cross-entropy loss integrated with attention mechanisms mitigates class imbalance without synthetic augmentation, preserving physiological integrity.Evaluated on the MIT-BIH Arrhythmia Database, DAN-ECGC achieves 98.45% accuracy, 98.70% precision, 98.88% weighted-F1 and 85.05% macro-F1, with 1.27M parameters and 20 ms inference time, ensuring strict real-time inference under fixed-window constraints and enabling efficient CPU-based deployment on embedded and wearable systems.

Fields of Study & Subjects

ECG Monitoring and Analysis Cardiac electrophysiology and arrhythmias Non-Invasive Vital Sign Monitoring

Publication details

Journal / source
International journal of intelligent engineering and systems
Volume
19
Issue
4
Pages
34–60
Publication date
23 March 2026
Language
EN