International Journal of Advanced Engineering Application

CNN-LSTM Deep Learning Framework for Multi-Class Cardiac Arrhythmia Detection from 12-Lead ECG with Wavelet Feature Extraction and HRV Analysis

Author(s):Hiroshi Nakamura

Affiliation: Department of Medical Engineering, Osaka University, Osaka, Japan

Page No: 43-46

Volume issue & Publishing Year: Volume 3, Issue 4, 2026/04/07

Journal: International Journal of Advanced Engineering Application (IJAEA)

ISSN NO: 3048-6807

DOI: https://doi.org/10.5281/zenodo.19478844

Download PDF Cite this article

Article Indexing:

Abstract:
Cardiac arrhythmias — abnormal electrical conduction patterns in the heart manifesting as irregular rhythm, aberrant waveform morphology, or disturbed rate — represent a leading cause of sudden cardiac death and constitute a major diagnostic and monitoring burden on India's overstretched cardiology services. The 12-lead ECG remains the gold standard for arrhythmia diagnosis, but the interpretation of continuous ECG recordings from Holter monitors (24-72 hours, generating 86,400-259,200 cardiac cycles) places demands on cardiologist time that are incompatible with health system capacity in a country where the cardiologist-to-population ratio is 1:250,000. This paper presents a CNN-LSTM hybrid deep learning model for automated multi-class arrhythmia detection from 12-lead ECG signals, incorporating continuous wavelet transform (CWT) spectrotemporal feature extraction and heart rate variability (HRV) power spectral analysis as supplementary feature channels alongside the raw ECG time-series. The model is trained on 34,200 ECG recordings from the PhysioNet Computing in Cardiology 2020 Challenge dataset augmented by 8,600 recordings from Apollo Hospitals Coimbatore's anonymised Holter archive, spanning five arrhythmia classes: Normal Sinus Rhythm, Atrial Fibrillation, Ventricular Tachycardia, Premature Ventricular Contractions, and Atrial Flutter. The CNN-LSTM achieves 96.2% overall accuracy, 95.8% macro-F1, and per-class AUC 0.986-0.994, outperforming SVM, kNN, Random Forest, standalone CNN, and standalone LSTM baselines.

Keywords: ECG, arrhythmia, deep learning, CNN-LSTM, atrial fibrillation, wavelet transform, HRV, cardiac monitoring, 12-lead, PhysioNet, biomedical signal processing, India

Reference:

• [1] Acharya, U. R., et al. (2017). A deep convolutional neural network model to classify heartbeats. Computers in Biology and Medicine, 89, 389-396.
• [2] Goldberger, A. L., et al. (2000). PhysioBank, PhysioToolkit, and PhysioNet. Circulation, 101(23), e215-e220.
• [3] IHRS. (2022). Prevalence of Cardiac Arrhythmias in India. Indian Heart Rhythm Society Annual Report.
• [4] Nakamura, H., & Tanaka, M. (2022). Compressed CNN for wearable arrhythmia detection. IEEE Transactions on Biomedical Circuits and Systems, 16(4), 612-624.
• [5] Rajpurkar, P., et al. (2017). Cardiologist-level arrhythmia detection with convolutional neural networks. arXiv:1707.01836.
• [6] Sundaram, M., & Rajan, K. (2023). Transfer learning for ECG arrhythmia across Indian and Japanese populations. Biomedical Signal Processing and Control, 84, 104813.
• [7] Task Force of ESC and NASPE. (1996). Heart rate variability: Standards of measurement. European Heart Journal, 17(3), 354-381.
• [8] Yıldırım, Ö., et al. (2018). Arrhythmia detection using deep convolutional neural network. Computers in Biology and Medicine, 102, 411-420.
• [9] Zheng, J., et al. (2020). A 12-lead ECG database to identify arrhythmias. Scientific Data, 7, 319.