Full text
Leveraging profiling to bridge healthcare silos for federated analyses Nelly Barret, Anna Bernasconi, Cinzia Cappiello, Giacomo Palu, and Pietro Pinoli Politecnico di Milano June 27th, 2025
Context: Federated Learning for healthcare 27/06/2025 Profiling healthcare silos - Nelly Barret et al. Which genes favor intelligence disability in children? How do genetic and clinical data correlate in eye dystrophies? 2/18
Context: Federated Learning for healthcare 27/06/2025 Profiling healthcare silos - Nelly Barret et al. Which genes favor intelligence disability in children? How do genetic and clinical data correlate in eye dystrophies? I-ETL [1] [1] N. Barret et al. I-ETL: an interoperability-aware health (meta)data pipeline to enable federated analyses. Under review. 3/18
Context: Federated Learning for healthcare 27/06/2025 Profiling healthcare silos - Nelly Barret et al. Which genes favor intelligence disability in children? How do genetic and clinical data correlate in eye dystrophies? I-ETL [1] [1] N. Barret et al. I-ETL: an interoperability-aware health (meta)data pipeline to enable federated analyses. Under review. [2] O. Beyan, et al. Distributed analytics on sensitive medical data: the personal health train. Data Intelligence. 2020 Jan 1;2(1-2):96-107. PHT [2] 4/18
Modeling and profiling healthcare silos 27/06/2025 Profiling healthcare silos - Nelly Barret et al. 5/18
Modeling and profiling healthcare silos 27/06/2025 Profiling healthcare silos - Nelly Barret et al. FAIRification 6/18
FAIRification in a nutshell For a given hospital: 27/06/2025 Profiling healthcare silos - Nelly Barret et al. Data Any patient data (real or synthetic) Metadata Any information providing context/semantic on the data I-ETL Extract Transform Load FAIR database 7/18 Our FAIR data model
Modeling and profiling healthcare silos 27/06/2025 Profiling healthcare silos - Nelly Barret et al. 8/18
Problem statement How to help users discover homogenized silos to design FL algorithms? We aim at a catalogue which: •General and adaptive •Discover data and metadata 27/06/2025 Profiling healthcare silos - Nelly Barret et al. 9/18
Building the global catalogue Combine individual catalogues of each silo: •Union of all the individual catalogues (one per hospital) •Easy thanks to our general catalogue model •Allows to implement visualisations for all centers at a time When hospitals update their data: •Recompute the underlying catalogue •Replace the old with the new individual catalogue 27/06/2025 Profiling healthcare silos - Nelly Barret et al. 16/18
Lessons learnt and our vision •Existing works: lack of generality, approach per use-case •FL requires highly homogenized infrastructures •Balance generalization vs. domain knowledge •Data and metadata are both necessary Abstracting current conceptual models is promising: → Reuse, use case-agnostic pipelines, various settings → Easy exploration of large and complex silos 27/06/2025 Profiling healthcare silos - Nelly Barret et al. 17/18
Thank you! 27/06/2025 Profiling healthcare silos - Nelly Barret et al. https://www.better-health-project.eu/ Next steps: silo template queries, LLM-assisted queries, … 18/18