Successfully implementing federated learning on real-world medical data from clinical routine requires significant effort due to variability in data quality, infrastructure, and compliance across institutions. A key factor in overcoming these challenges is the adherence to standardized protocols across clinics, which simplifies data integration. In imaging studies, the inherent matching capabilities of DICOM provide a powerful framework for aligning and processing data across various modalities, enhancing interoperability. To ensure accessibility and usability for medical professionals without a computer science background, a well-designed graphical user interface (GUI) is beneficial. A user-friendly dashboard simplifies interaction with federated learning systems, enabling clinicians and researchers to focus on medical insights rather than technical complexities.

Within the FLOTO project of the Deutsche Zentrum für Herz-Kreislauf-Forschung (DZHK), we demonstrate the effectiveness of our approach through two successful federated learning applications:

1. Federated Knowledge Distillation for Segmentation and Point Detection on CT data, which enhances model performance despite small amount of labelled data while preserving data privacy.

2. Multimodal Risk Prediction for Pacemaker Implantation after Heart Valve Replacement, showcasing the potential of federated learning in clinical decision support.