Leveraging Profiling to Bridge Healthcare Silos for Federated Analyses

Healthcare is more and more relying on digital information, bringing new challenges for its management, exploration, and usage. Healthcare data represents a challenge for information systems because, for privacy regulations, it cannot exit the original silo in which it has been produced (typically owned by hospitals), and may be of various kinds (clinical reports, DNA sequences, MRI scans, etc.). To manage this complexity, it is natural to use Federated Learning to safely analyze the underlying silos’ content. However, designing and running federated algorithms requires to know what the silos contain and how they can be joined (on which common attributes). Existing catalogs provide prelim-inary visualizations, which are hardly generalizable due to their under-lying use-case-tailored data models. To overcome these limitations, we provide a general catalog conceptual model as well as profiling techniques to extract information of interest from silos. Our proposed catalog is gen-eral enough to be used in various healthcare scenarios with diverse kinds of data. It also facilitates experts’ work in creating Federated Learning algorithms running in networks of interoperable healthcare silos.