Medical Image Classification using Hybrid Quantum-Classical Neural Network

Abstract

Medical image classification plays a crucial role in healthcare, enabling accurate

disease diagnosis and treatment planning. Traditional deep learning models,

such as Convolutional Neural Networks (CNNs), have demonstrated remarkable

success in this domain but are often constrained by high computational costs,

data inefficiencies, and challenges in generalization, particularly for small and

imbalanced medical datasets. Quantum computing has emerged as a promising

alternative, offering computational advantages through quantum parallelism

and high-dimensional feature representations. This project explores the integration

of quantum computing with classical deep learning to develop a Hybrid

Quantum-Classical Neural Network (HQCNN) for binary medical image classification.

The proposed model leverages a classical CNN for feature extraction

and a quantum circuit for enhanced feature transformation, optimizing classification

performance while reducing computational overhead. The model is

evaluated on four benchmark medical datasets: BloodMNIST, OrganAMNIST,

and PathMNIST. Experimental results demonstrate that the HQCNN model

achieves 99.12% accuracy and 99.91% AUC on BloodMNIST, 98.84% accuracy

and 99.00% AUC on OrganAMNIST, and 97.83% accuracy with 100.00% AUC

on PathMNIST, outperforming traditional CNNs in most cases. While the

proposed HQCNN model enhances classification accuracy and robustness, challenges

such as quantum noise, limited qubit scalability, and training convergence

issues remain significant obstacles. Future research will focus on optimizing quantum circuit architectures, improving hybrid training strategies, and scaling the model for high-resolution medical imaging applications. This study highlights the potential of quantum-assisted deep learning in medical imaging and paves the way for future advancements in quantum-based healthcare solutions