Federated ECG Classification with Digital Twin Orchestration and Entropy-Based Client Selection

Digital healthcare (eHealth) is a key vertical for future sixth-generation networks (6G), especially in the wake of the Covid-19 pandemic, which emphasized the need for widespread telemedicine solutions. The advent of 6G, thanks to the technologies that will characterize it, offers several growth opportunities for the eHealth sector. For example, artificial intelligence (AI) is currently widely used in healthcare. However, the privacy of data and information is not always guaranteed by the network architectures that support such applications. In this paper, we present a novel federated learning (FL) framework that combines digital twin (DT) simulation and edge intelligence to enable distributed learning while preserving data privacy. We demonstrate this by simulating a realistic edge-learning scenario in which electrocardiogram (ECG) data from patients are transmitted to distributed virtual twins, and show that selectively including only the most informative clients based on prediction uncertainty significantly improves both convergence speed and model generalization. Specifically, the proposed DT strategy outperforms both conventional and purely entropy-based approaches on key evaluation measures by using past trends and local training performance to predict future client utility. The proposed architecture provides a versatile and flexible foundation for implementing AI-driven and privacy-friendly healthcare systems.