A Hybrid Quantum–Classical Generative Adversarial Network for Synthetic Medical Data Generation and Disease Diagnosis

Abstract

Medical image analysis, as a crucial application of AI, plays a pivotal role in supporting early disease detection and diagnosis. One of the biggest problems in medical imaging is the lack of adequate and good quality datasets, particularly for disease classes that are more critical like malignant cases. Classical machine learning algorithms generally do not generate good generalization and the algorithms are not sensitive on imbalanced datasets. In this paper, to solve this problem, a hybrid quantum-classical generative adversarial network (QGAN) framework is designed and analyzed for synthetic medical data generation and enhanced disease classification. Conventional convolutional neural networks (CNNs) were enhanced by class-weighted learning and hybrid quantum-classical generative model (QCGM) using the HAM10000 skin lesion dataset, using PennyLane and TensorFlow. From experimental observation, the hybrid quantum-classical augmentation approach shows a considerable improvement in case detection of malignancy over the conventional approach. Lastly, a prototype medical web application was built using AI text-to speech for practical implementation.