Federated Digital Twin Architecture with Synthetic Data Generation for Privacy-Preserving Diabetes Management

Abstract

Diabetes mellitus management via AI and other ICT solutions suffers from data scarcity and privacy constraints. This paper proposes a unified approach that addresses both challenges by integrating AI-driven synthetic data generation with a Digital Twin (DT) framework, built on a privacy-preserving federated learning and decentralised data storage architecture. We present an end-to-end system where each patient’s health data is securely stored in a personal data pod under their control, and a global predictive model is trained collaboratively across pods without centralizing raw data. To overcome limited real-world data, we incorporate synthetic data generation techniques (e.g. generative neural networks) to augment training datasets and simulate diverse clinical scenarios in a GDPR-compliant manner. We detail the system design, including the synthetic data pipeline, federated model training process, and DT-driven prevention strategies. Experimental considerations demonstrate that our approach can achieve predictive performance comparable to centralized models while strictly preserving privacy.

Ähnliche Arbeiten