International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 13 Issue: 06 | Jun 2026
p-ISSN: 2395-0072
www.irjet.net
Quantum Machine Learning for Cardiovascular Disease Prediction: A Comprehensive Comparative Study of Classical and Quantum Approaches Across Multi-Modal Cardiac Datasets Varanasi Kushal1, Dr. Y. Anuradha2, Dr. P Krishna Subba Rao3 1,2,3Gayatri Vidya Parishad College of Engineering (Autonomous), Visakhapatnam, India
------------------------------------------------------------------------------------***------------------------------------------------------------------------------------
1. INTRODUCTION
Abstract–Cardiovascular disease (CVD) has become the leading cause of death in the world, accounting for an estimated 17.9 million deaths per year, or roughly one-third of all deaths globally. With accurate and timely detection, this toll can be substantially reduced, yet a scalable, interpretable, and clinically reliable prediction system remains largely unrealized. This study presents an in-depth comparison of eight classical and two quantum machine learning models across four cardiovascular datasets comprising 44,645 clinical records, ECG images, and physician-verified signals. While classical methods such as LightGBM achieved 75.3% accuracy and a 0.832 ROC-AUC, the best-performing quantum model, the Hybrid Quantum Multi-Layer Perceptron (HQMLP), reached only 57.0% accuracy under present-day simulated computing constraints. Notably, perfect classification was achieved on a small, high-quality sample of the PTB-XL ECG dataset, indicating that label accuracy and data normalization matter more than sheer data volume. Using an integrated twelve-dimensional classical-quantum evaluation framework that considers classification performance, time complexity, robustness to noise, training time, and clinical explainability, this work demonstrates that classical ML has reached a stage suitable for clinical deployment, evidenced by an 87.5/100 deployment-readiness score, throughput exceeding 200 queries per second over REST, and a 100–300x computational speed advantage over quantum approaches on current hardware. The top predictive features were chest pain type (15.2%), maximum heart rate (12.8%), and ST-segment depression (11.5%), consistent with the established pathophysiology of coronary artery disease. These findings establish a framework for evaluating deployment readiness in cardiovascular ML while highlighting the aspects of quantum hardware development most critical to future medical applications.
Cardiovascular disease presents the greatest disease burden globally. The World Health Organization estimated 17.9 million premature deaths from CVD in 2004, occurring predominantly in developing regions where specialized clinical skills and diagnostic equipment remain scarce. The majority of cardiac morbidity myocardial infarction, sudden cardiac death, and related events is preceded by identifiable physiological warning signs. If these early indicators could be reliably extracted from clinical data before irreversible physical damage occurs, timely intervention becomes possible. Machine learning offers a scalable and costeffective means of achieving this. The evolution of machine learning for medical diagnosis can be traced from simple decision trees and logistic regression models trained on the UCI Heart Disease dataset, through complex ensemble methods that substantially improved predictive power, to neural networks trained directly on ECG waveforms and cardiac imaging data. The emergence of gradient boosting libraries XGBoost, LightGBM, and CatBoost enabled these methods to outperform neural networks on tabular medical data while also offering improved interpretability and faster training. At the same time, the digitization of healthcare records and the broader sharing of large ECG datasets has made a growing variety of cardiac data more accessible to researchers. Quantum machine learning (QML) is an emerging theoretical extension of classical ML that exploits quantum superposition, entanglement, and interference to encode exponentially large feature spaces using only a polynomial number of parameters. Quantum kernel methods are theorized to offer learning advantages over classical kernels in high-dimensional regimes. While these theoretical benefits are compelling, the empirical utility of QML in clinical practice remains largely unassessed, constrained by the limitations of current quantum hardware and classical simulators.
Key Words: Quantum Machine Learning, Cardiovascular Disease Prediction, Gradient Boosting, LightGBM, Hybrid Quantum MLP, ECG Signal Classification, Clinical Decision Support, Feature Importance
This work addresses the absence of a systematic comparison between classical and quantum ML that spans
© 2026, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 1013