Hybrid CNN-transformer networks with cross-representation attention for ECG-based cardiovascular disease classification
Journal of Science and Technology - HaUI · 30.03.2026 · 62 (3E)
DOI: 10.57001/huih5804.2026.080 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Science and Technology - HaUI, online erschienen am 30.03.2026. Verfasst von Anh Dung Ho.
Abstract (Originalsprache)
Electrocardiogram (ECG) analysis plays an important role in the early diagnosis of cardiovascular diseases.However, the complexity of ECG waveforms presents significant challenges for automated classification.This paper proposes a novel Hybrid CNN-ViT framework for cardiovascular disease classification, where one-dimensional ECG signals are transformed into three complementary two-dimensional representations: Gramian Angular Field (GAF), Recurrence Plot (RP), and Markov Transition Field (MTF).The proposed framework combines CNN-based local feature learning and Transformer-based global contextual modeling through an Adaptive Attention Fusion mechanism to effectively capture complementary ECG characteristics.In addition, a Cross-Channel Attention (CCA) mechanism is introduced to model inter-representation dependencies and enhance discriminative feature interactions among GAF, RP, and MTF representations.Experiments conducted on the PTB-XL dataset demonstrate that the proposed Hybrid CNN-ViT framework achieves a sensitivity of 89.12% and a specificity of 80.38%, outperforming conventional CNN-based models, Transformer-based models, and a recent state-of-the-art (SOTA) method.The results confirm that combining multiple 2D ECG representations with Hybrid CNN-ViT learning and Cross-Channel Attention significantly improves cardiovascular disease classification performance while maintaining a balanced trade-off between sensitivity and specificity, which is critical for biomedical diagnostic applications.
Zitieren
APA
Ho, A. D. (2026). Hybrid CNN-transformer networks with cross-representation attention for ECG-based cardiovascular disease classification. Journal of Science and Technology - HaUI, 62(3E). https://doi.org/10.57001/huih5804.2026.080
Vancouver
Ho AD. Hybrid CNN-transformer networks with cross-representation attention for ECG-based cardiovascular disease classification. Journal of Science and Technology - HaUI. 2026;62(3E). doi:10.57001/huih5804.2026.080
BibTeX
@article{Ho2026,
author = {Ho, Anh Dung},
title = {Hybrid CNN-transformer networks with cross-representation attention for ECG-based cardiovascular disease classification},
journal = {Journal of Science and Technology - HaUI},
year = {2026},
volume = {62},
number = {3E},
doi = {10.57001/huih5804.2026.080}
}
EUVIMED