TCNA: A Temporal Convolutional Network With Attention for Real-Time Multi-Class ECG Classification at the Edge
IEEE Access · 2026 · 14 · 96191-96207
DOI: 10.1109/access.2026.3706663 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Aicha Aggoune. Die Arbeit nennt 45 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The electrocardiogram (ECG) is a non-invasive signal of the heart’s electrical activity, widely used for detecting abnormalities. Despite the development of numerous methods for automatically classifying heartbeats, their performance remains limited by insufficient extraction of high-level features, particularly in real-time, where accuracy degrades significantly. We propose a real-time ECG framework based on TCNA, a dual-path model that combines deep temporal representations retrieved by the Temporal Convolutional Network-attention branch with clinically important ECG interval data (PR, ST, QT, and TP segments). By combining data-driven temporal patterns with domain-specific cardiac descriptors, the model captures both long-range signal dependencies and cardiac cycle physiological properties. An interactive learning is provided using Grafana dashboards and InfluxDB time series database, allowing domain experts to validate and label anomalies in real time, creating a feedback loop that improves classification accuracy over time. Thorough experiments are based on two validation modes: offline validation using PTB-XL benchmark dataset and online real-time validation using ECG signals acquired from edge devices. The model achieves an accuracy of 99.56%, precision of 98.67%, recall of 99.80% for offline mode and for real-time mode with accuracy of 99.71%, a precision of 98.87%, and recall of 99.91%. These results confirm the effectiveness, robustness, and practical suitability of TCNA for ECG monitoring applications.
Zitieren
APA
Aggoune, A. (2026). TCNA: A Temporal Convolutional Network With Attention for Real-Time Multi-Class ECG Classification at the Edge. IEEE Access, 14, 96191-96207. https://doi.org/10.1109/access.2026.3706663
Vancouver
Aggoune A. TCNA: A Temporal Convolutional Network With Attention for Real-Time Multi-Class ECG Classification at the Edge. IEEE Access. 2026;14:96191-96207. doi:10.1109/access.2026.3706663
BibTeX
@article{Aggoune2026,
author = {Aggoune, Aicha},
title = {TCNA: A Temporal Convolutional Network With Attention for Real-Time Multi-Class ECG Classification at the Edge},
journal = {IEEE Access},
year = {2026},
volume = {14},
pages = {96191--96207},
doi = {10.1109/access.2026.3706663}
}
EUVIMED