DiaBeCo: A Decentralized Federated Learning Architecture for Diabetes Care with Personal Data Pods and Digital Twin

Abstract

Diabetes mellitus is a widespread chronic disease that requires continuous monitoring and personalized treatment. Traditional predictive analytics often rely on aggregating sensitive patient data in centralized servers, raising privacy concerns and conflicting with modern data protection regulations. This paper introduces DiaBeCo, a patient-centric architecture that unifies solid personal data pods, Federated Learning (FL) and Digital Twin (DT) to enable privacy-preserving diabetes care. In DiaBeCo, each patient's health data resides in their solid pod under their control, and machine learning models are trained collaboratively using FL so that no raw data ever leaves a pod. The resulting global model drives an individualized DT for each patient, which allows 'what if' simulations of interventions (e.g., changes in diet or activity) to produce personalized risk projections. By design, only non-identifying model updates and aggregate information are shared; while, all personal health records remain encrypted in the patient's pod, with consent governed by the patient. This integration empowers patient data ownership, improves predictive accuracy through pooled learning, and yields personalized decision support - all while complying with stringent privacy regulations. DiaBeCo demonstrates a novel approach to decentralized, trust-centric digital healthcare by combining Solid pods, FL, and DT for enhanced diabetes management.

Ähnliche Arbeiten