From Distributed Data to Central Models: Federated Machine Learning in Medicine (XtraTrack Season 2026)

10-06-2026 16:00

Routine clinical data collected in hospitals holds great potential for use in clinical decision support and medical research. Modern machine learning methods can create predictive models for complex clinical questions across a wide range of medical domains. In practice, however, the use of such data is severely limited, not only to legal and ethical constraints, but also to organisational and technical challenges. Federated learning (FL) represents one possible approach to address some of these limitations. It enables decentralized training and evaluation of machine learning models, while sensitive, high-resolution patient data remains at local institutions. Furthermore, it has the potential to produce more generalisable models, as it can be trained on different data in various institutions. Despite these advantages, federated approaches introduce additional challenges. From a technical perspective, they often require increased expertise, infrastructure and time at the participating sites. From an organisational and regulatory standpoint, federated analysis is still insufficiently understood. From a methodological standpoint, it remains an open question how federated models should be optimally trained and evaluated, and under which conditions they can outperform or complement centrally or locally trained models, incorporating the benefits of diverse datasets from various clinics. This talk provides an overview of federated analysis and learning approaches in medicine, illustrating their application through selected examples from intensive care practice and highlighting their potential to advance collaborative, multi-center research.

PD Dr. rer. nat. Mathias Kaspar - Institute of Digital Medicine, Augsburg University Hospital

PD Dr. rer. nat. Mathias Kaspar is a senior researcher in medical informatics with extensive experience in both academia and industry. He completed his studies and PhD in Medical Informatics at the Georg-August University Göttingen, including a two-year research stay at the Computation Institute, University of Chicago. He gained industry experience at Siemens Health in Erlangen and the USA. Dr. Kaspar has held long-term research positions at the German Center for Heart Failure in Würzburg, the Department of Health Services Research at Carl von Ossietzky University Oldenburg, and the Institute for Digital Medicine. His research focuses on the provision and utilization of routine medical data for research, including data preparation, enabling new use cases, and big data analytics using statistics, machine learning, and federated analysis.

Organized by GMDS AG MoCoMed / Impressum / Privacy