(126g) Quantum-Enhanced Federated Learning for Secure and Efficient Biomedical Image Classification

Advancements in artificial intelligence (AI) are transforming biomedical imaging by enabling early disease detection and accurate diagnostics. However, stringent privacy regulations such as HIPAA and GDPR impose significant constraints on data sharing among healthcare institutions.¹ Federated learning (FL) has emerged as a promising solution by facilitating decentralized model training without requiring direct data exchange.^2,3 Despite its advantages, FL remains susceptible to adversarial threats, including model inversion and privacy leakage.⁴

This study introduces a quantum federated learning (QFL) framework that integrates FL with fully homomorphic encryption (FHE) and quantum computing to enhance data security, computational efficiency, and scalability.⁵ Unlike conventional FL, where model updates may inadvertently expose sensitive patterns, our approach leverages FHE to ensure that all computations occur on encrypted data, eliminating the risk of information leakage.^6,7 However, the computational burden of encrypted training presents a major challenge, which we address by incorporating quantum machine learning (QML).^5,8 By integrating quantum-enhanced optimization techniques, we mitigate the performance trade-offs associated with encrypted federated training, leading to improved efficiency and robustness.^9,10

Our framework is particularly well-suited for applications in medical imaging, such as MRI and CT scan analysis, where AI-driven models play a crucial role in diagnostics and treatment planning.¹¹ Additionally, we extend our methodology to multitopic data integration, demonstrating the potential of variational quantum privacy-preserving FL in complex biomedical applications.¹² By harnessing the power of quantum computing and advanced encryption methods, this work provides a scalable, privacy-compliant, and high-precision AI framework for next-generation medical imaging solutions.