Enhanced federated learning for secure medical data collaboration

Abstract

Abstract Federated learning (FL) enables collaborative model training across multiple institutions while preserving data privacy. However, conventional encryption techniques used in FL remain vulnerable to quantum attacks, raising concerns about the security of model update transmissions. To address this, we propose a quantum-safe federated learning framework, which integrates a post-quantum lattice-based digital signature scheme into the FL communication protocol. Unlike traditional cryptographic methods, lattice-based signatures provide resilience against quantum adversaries while maintaining efficiency in resource-constrained environments. Our approach is evaluated on brain tumor and colorectal polyp segmentation datasets from Kaggle, demonstrating its effectiveness in securing FL updates with minimal computational overhead. The results highlight the feasibility of quantum-secure FL, particularly in hospital environments with limited quantum computing access.