A large scale statistical analysis of quantum and classical neural networks in the medical domain

TL;DR

This paper compares classical and quantum neural networks for predicting heart disease, finding that quantum models perform well even with limited data.

Contribution

The study introduces a structured methodology for evaluating quantum neural networks in medical prediction tasks.

Findings

Quantum Neural Networks (QNNs) achieve comparable accuracy to classical models in heart disease prediction.

QNNs show potential advantages in data-scarce scenarios, which are common in clinical settings.

A reproducible evaluation framework for QNNs is presented, focusing on key design parameters.

Abstract

Classical neural networks (NNs) have shown strong performance in medical data analysis. However, they typically require large labeled datasets and may struggle in data-scarce scenarios, common in clinical practice. Quantum Neural Networks (QNNs) have emerged as a promising alternative. This paper presents a comparative study between NNs and QNNs for heart disease prediction, addressing the limitations of current models in low-data regimes. We systematically evaluate 460 QNNs (using 11-13 qubits) and 4,480 NN architectures, analyzing key design parameters: encoding schemes, re-uploading strategies, circuit depth, and dropout (for QNNs), as well as hidden layers, neurons per layer, and dropout (for classical NNs). Top-performing models are selected for a direct comparison in terms of accuracy and sample complexity. Our results show QNNs achieve comparable accuracy and demonstrate potential advantages in data-scarce settings. Our study presents a structured and reproducible methodology for evaluating QNNs in clinical contexts, thereby supporting the broader investigation of quantum machine learning in applied healthcare domains.

Figures