Explorando a Viabilidade da Computação Quântica em Dados Biomédicos: Um Benchmarking entre Processamento Digital Clássico e Algoritmos Quânticos

Abstract

The computational analysis of biomedical data involves noisy, non-stationary, and often nonlinear signals. Among these signals, Heart Rate Variability, obtained from electrocardiogram-derived RR interval time series, provides relevant information about autonomic regulation of cardiac activity and may support the distinction between physiological patterns associated with normotensive individuals and individuals with hypertension under pharmacological intervention. This work investigates the feasibility of Quantum Machine Learning algorithms in this context through a benchmark between classical digital processing and simulated hybrid quantum models. The methodology includes the extraction of temporal, entropic, fractal, geometric, and spectral descriptors, the construction of a classical baseline using ensemble models, and the development of quantum architectures based on Quantum Support Vector Classifier and Quantum Feature Maps. The study compares consolidated experimental milestones obtained after successive engineering tests, including variations in angular encoding, feature selection, circuit depth, entanglement topology, and Data Re-uploading. The results indicate that the classical reference model achieved a global mean accuracy of 83.73% and a peak accuracy of 91.67%, while the most stable hybrid quantum configuration achieved a mean accuracy of 80.17% and a peak accuracy of 90.00% in ideal simulation, using only four main descriptors redistributed across eight qubits. The findings suggest that the quantum approach does not currently replace mature classical methods, but it shows experimental competitiveness and dimensional efficiency in a realistic biomedical problem based on cardiac signals. The results should be interpreted as exploratory evidence obtained from a limited-scale dataset, rather than as definitive clinical validation. The research also highlights relevant limitations of noisy intermediate-scale quantum computing, such as simulation cost, encoding restrictions, noise sensitivity, and dependence on rigorous classical preprocessing.