Trustworthy AI for Stroke Pathways improvement.
Funding: HORIZON-HLTH-2022-STAYHLTH-01-two-stage. 2023-2027.
Partners: Fundacio Hospital Universitari Vall D’Hebron – Institut de Recerca, Fondazione Policlinico Universitario Agostino Gemelli-IRCSS, KU Leuven, UKC Ljubljana, EATRIS, CERN, EURECAT, JOZEF STEFAN Insitute, NORA Health, STROKE ALLIANCE FOR EUROPE, Consiglio Nazionale delle Ricerche, Politecnico di Milano, NACAR.
The project proposes a novel trustworthy by design and privacy-preserving AI-based platform to assist clinicians, patients and caregivers in the management of acute and chronic phases of ischemic stroke, based on the integration of clinical and patient-reported data for a trustworthy assessment of disease progression. A Federated Learning (FL) infrastructure (hosted by CERN) will be developed and used to enable multiple entities to collaboratively build several trustworthy AI based predictive models by leveraging stroke data without compromising privacy and implementing best-in-class security protocols, easing compliance with GDPR and lowering adoption barriers by hospitals.
FL training pipelines and the underlying network developed by CERN, CNR and Politecnico di Milano, will be adapted for its unprecedented use in healthcare. The partners will work closely with the appointed IT teams and Data Scientists at each participant hospital, to set-up the local infrastructure at the clinical nodes. The main concepts of the proposed architecture are depicted in the figure: local AI models are trained independently by each node (i.e., medical site/institution) of the network and are shared with a Parameter Server (PS) that produces, after some iterations, an augmented federated model, the global AI model, thanks to the multi-institution collaboration. Client-server and decentralised network architectures are defined and optimised to orchestrate the FL process. For all cases, patient data and personal information are kept on each individual node and never transferred over the network. Finally, the developed TRUSTroke PS infrastructure will be hosted and operated at CERN, by a secure and robust computing environment.
More info on: https://trustroke.eu/