A Quantum-Inspired Deep Learning Models for Skin Lesion Classification

The present study explores the application of quantum machine learning for classifying skin lesion images as melanoma, nevus, or basal cell carcinoma. Using hybrid classical-quantum neural networks with transfer learning, we demonstrate the potential advantages of the four proposed variants in generalizing on complex data. Our models yield improved outcomes compared to conventional deep learning models such as ResNet-18, DenseNet-121, RegNetX-1.6GF, and EfficientNetV2-small. Specifically, we achieve accuracy rates of 82.04%, 85.0%, 84.90% and 86.94% as opposed to 80.0%, 84.08%, 81.63% and 84.49%, respectively. The experiments are conducted on the default PennyLane simulator using the ISIC dataset, demonstrating the practical applicability of quantum machine learning for computer-aided diagnosis applications. Despite these encouraging findings, the study highlights the need for further research to comprehensively understand quantum machine learning approaches and optimize their performance.