Real-Time Intelligent Healthcare Enabled by Federated Digital Twins With AoI Optimization
Beining Wu, Jun Huang, Qiang Duan
Abstract
Meeting the demand for prompt yet accurate diagnosis remains a major challenge in intelligent healthcare, particularly under conditions involving emergencies or rare medical cases with insufficient data. This article reviews existing AI-driven methods in healthcare and identifies their strengths and limitations in addressing these challenges. To tackle the urgent need for timely medical responses, we propose a novel federated digital twin framework incorporating split learning and artificial intelligence-generated content (AIGC). Additionally, we leverage Age of Information (AoI) as a comprehensive metric to optimize real-time data and model updates, employing a deep reinforcement learning algorithm to achieve enhanced information freshness. Our framework exhibits high diagnostic accuracy across diverse clinical datasets, including Brain MRI (93%), OrganA MNIST (89%), Chest MNIST (83%), and Retina MNIST (80%). Moreover, experimental validation demonstrates our method significantly reduces the AoI to an average of 0.6s, notably outperforming Rainbow DQN and traditional greedy approaches. These results unveil the potential of our proposed approach to deliver timely, privacy-preserving, and accurate diagnostics, laying a foundation for future advancements in intelligent healthcare systems.