Design and Implementation of a Deep Learning–Driven ECG Signal Classification System for Cardiac Arrhythmia Detection

View at publisher Summarise with VIRAT
Authors: A. Gnana Rooba, R Prem Ananth, J. Relin Francis Raj, P. Bharathi, J. Nirmal Jothi, N Soundiraraj
Cited by: 0
References: 15

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

Automated electrocardiogram (ECG) analysis represents a significant engineering problem due to the need for reliable signal interpretation under noisy conditions, limited labeled data, and high inter-patient variability. From an engineering perspective, the development of robust computational models with scalable technological capabilities is essential for next-generation cardiac monitoring systems. This work presents a deep learning–based engineering framework for automated arrhythmia detection that combines unsupervised feature learning with supervised classification. A convolutional denoising autoencoder is employed to learn compact and noise-resilient representations directly from raw ECG signals, reducing dependence on handcrafted features and enabling systematic Scientific research in ECG signal processing. The learned representations are subsequently integrated with a hybrid classifier composed of bidirectional long short-term memory networks and an attention mechanism to model temporal dependencies and morphological variations in cardiac rhythms. The proposed framework was implemented using a publicly available benchmark ECG dataset and evaluated under a patient-wise experimental protocol to ensure realistic performance assessment. The system demonstrates strong generalization capability and improved classification accuracy compared to conventional end-to-end supervised deep learning models. From an engineering systems viewpoint, the integration of attention-based interpretability supports deployment in real-time decision-support platforms built on modern ICT infrastructure and scalable computing Infrastructure. The results highlight the applicability of advanced deep neural architectures in ECG signal classification tasks and emphasize the role of algorithmic optimization and model design in improving diagnostic reliability. Overall, the proposed approach reflects meaningful Innovation in ECG signal analysis by unifying unsupervised representation learning, temporal modeling, and explainable artificial intelligence within a single engineering framework suitable for embedded and wearable health monitoring applications.

Fields of Study & Subjects

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

Publication details

Journal / source
International Conference on Image Processing and Capsule Networks (ICIPCN)
Pages
1750–1758
Publication date
27 January 2026
Language
EN