Recent advances in use of digital twins for healthcare applications have enabled real-time treatment adaptation and medical management through ML-driven predictions. In this paper, we design a digital twin system for medical/clinical applications that consists of real-time patient-specific physiological replicas, featuring dynamically rendered 3D organ models and ML analytics. These clinical twins are located on secure on-premises servers within hospital networks, ingesting sensor data to animate organ visualizations. In our framework, this Digital Twin system is integrated with a medical Metaverse. This allows physicians to access a unified virtual environment where spatially organized holograms of all patients' organs are streamed in real-time to VR headsets, enabling holistic clinical assessment. By means of the twins, it will be possible to exploit ML analytics for physiological signal prediction and anomaly detection. However, sensor-generated data volumes and privacy constraints rule out centralized cloud training. To address this, we introduce a cloud-based orchestrator that enables distributed model training while preserving data locality. We investigate and compare privacy-preserving distributed learning techniques, such as federated learning (FL) and split learning (SL). In order to make the distributed learning efficient in terms of communication resources, we weigh performance and novelty contribution to the learning process. Results prove the feasibility and efficiency of the proposed solution, together with practicality of the methodology to be exploited by physicians to quickly check patients conditions.

Metaverse-Enabled Digital Twins employing Distributed Learning for Clinical applications

Galluccio L.
Ultimo
Conceptualization
2025-01-01

Abstract

Recent advances in use of digital twins for healthcare applications have enabled real-time treatment adaptation and medical management through ML-driven predictions. In this paper, we design a digital twin system for medical/clinical applications that consists of real-time patient-specific physiological replicas, featuring dynamically rendered 3D organ models and ML analytics. These clinical twins are located on secure on-premises servers within hospital networks, ingesting sensor data to animate organ visualizations. In our framework, this Digital Twin system is integrated with a medical Metaverse. This allows physicians to access a unified virtual environment where spatially organized holograms of all patients' organs are streamed in real-time to VR headsets, enabling holistic clinical assessment. By means of the twins, it will be possible to exploit ML analytics for physiological signal prediction and anomaly detection. However, sensor-generated data volumes and privacy constraints rule out centralized cloud training. To address this, we introduce a cloud-based orchestrator that enables distributed model training while preserving data locality. We investigate and compare privacy-preserving distributed learning techniques, such as federated learning (FL) and split learning (SL). In order to make the distributed learning efficient in terms of communication resources, we weigh performance and novelty contribution to the learning process. Results prove the feasibility and efficiency of the proposed solution, together with practicality of the methodology to be exploited by physicians to quickly check patients conditions.
2025
Metaverse
Digital twins
Human organs
Distributed Learning
Federated Learning
Split Learning
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/20.500.11769/738790
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 2
  • ???jsp.display-item.citation.isi??? 0
social impact