Federated Learning for Securing Medical Imaging Against Deepfakes in 6G Smart Hospitals
Emma Perales, Romain Verdy-Ricard, Mohamed Aymen Labiod, Gueltoum Bendiab, Yasmina Chenoune · 2025
The convergence of $\mathbf{6 G}$ connectivity, artificial intelligence, and smart healthcare infrastructure is transforming medical imaging workflows. In 6G-enabled smart hospitals, radiology images such as CT and MRI scans are processed and transmitted at ultra-low latency between edge devices, cloud platforms, and remote specialists. However, this speed and openness introduce critical security vulnerabilities, notably the risk of deepfakebased manipulation. Malicious actors can exploit generative adversarial networks (GANs) to inject, erase, or alter anomalies in medical scans-potentially leading to misdiagnoses, insurance fraud, or patient endangerment. In this paper, we introduce a federated deepfake detection framework tailored to 6 G healthcare networks. Our approach uses a DenseNet-based convolutional neural network collaboratively trained across multiple medical institutions, ensuring privacy preservation through decentralised learning. We simulate edge-based federated training scenarios on benchmark datasets, including real and manipulated CT images, and evaluate the model’s robustness in both intra- and cross-dataset conditions. The results show strong generalisation capabilities and high sensitivity to tampered medical content. This work paves the way for a secure AI-assisted image verification pipeline in next-generation healthcare environments, reinforcing 6 G network management with decentralised, privacyaware intelligence for deepfake mitigation.