Citation

Alam, Touhidul and Aziz, Shusmita Anjum and Rahman, Sayedur and Abdullah-Al-Jubair, Md. and Nandi, Dip and Hossen, Md Jakir (2026) ECG-HeartNet: An Explainable Hybrid 1D CNN-BiLSTM Network for Accurate Cardiac Arrhythmia Diagnosis. In: 2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN), 16-18 April 2026, Chittagong, Bangladesh.

Text 5.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

Accurate and interpretable automated diagnosis of cardiac arrhythmias is critical for mitigating the global burden of cardiovascular diseases. While Deep Learning (DL) models often achieve high classification performance, their “blackbox” nature hinders clinical trust. This study proposes ECGHeartNet, a lightweight hybrid architecture combining 1D CNNs for morphological feature extraction and BiLSTM units for capturing temporal dependencies. Evaluated on the MIT-BIH Arrhythmia Dataset, the model achieves a state-of-the-art test accuracy of 98.60 % and a weighted F1-score of 98.61 %, significantly outperforming complex transfer learning architectures such as ResNet1D and VGG1D. Beyond performance, this study specifically addresses the interpretability challenge by integrating XAI frameworks. Through SHAP and LIME, we visualize the model's decision-making process, demonstrating that it correctly prioritizes clinically relevant features like the QRS complex over noise. The proposed framework offers a robust, high-accuracy, and transparent solution suitable for real-world deployment in cardiac monitoring systems.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Electrocardiogram (ECG), Arrhythmia Classification, Deep Learning, LSTM, CVD, XAI, SHAP, LIME
Subjects:	R Medicine > R Medicine (General) > R855-855.5 Medical technology
Divisions:	Faculty of Engineering and Technology (FET)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	04 Aug 2026 04:50
Last Modified:	04 Aug 2026 04:50
URII:	http://shdl.mmu.edu.my/id/eprint/16489

Downloads

Edit (login required)