QMicroNet: a quantum hybrid approach to robust cross-domain medical image analysis

Abstract

Lightweight deep learning models are essential for deploying medical image analysis systems on devices with limited resources. Nevertheless, many existing lightweight architectures experience underfitting and performance degradation in multi-domain scenarios. In response to these challenges, QMicroNet is introduced as a lightweight classical-quantum hybrid architecture for multi-domain medical image analysis. Specifically, QMicroNet integrates MicroNet, a lightweight convolutional neural network optimized for efficient feature extraction, with quantum neural networks (QNNs), thereby leveraging Hilbert-space properties for enhanced feature representation while maintaining a compact model size. According to experimental results on the RadImageNet dataset, which contains 1.35 million images, QMicroNet achieves superior Top-1 and Top-5 accuracies compared to traditional convolutional neural network (CNN) architectures, despite utilizing only 0.32 million parameters. Overall, these findings underscore the potential of quantum hybrid models to improve accuracy, generalizability, and feasibility in resource-constrained medical image analysis systems.