The radiological cooperative network (RACOON) is dedicated to strengthening Covid-19 research by establishing a standardized digital infrastructure across all university hospitals in Germany. Using a combination of structured reporting together with advanced image analysis methods, it is possible to train new models for a standardized and automated biomarker extraction that can be easily rolled out across the consortium. A major challenge consists in providing generic and robust tools that work well on relevant data from all hospitals, not just on those where the model was originally trained. Potential solutions are federated approaches that incorporate data from all sites for model generation. In this work, we therefore extend the Kaapana framework used in RACOON to enable real-world federated learning in clinical environments. In addition, we create a benchmark of the nnU-Net when applied in multi-site settings by conducting intra- and cross-site experiments on a multi-site prostate segmentation dataset [1].
CITATION STYLE
Kades, K., Scherer, J., Zenk, M., Kempf, M., & Maier-Hein, K. (2023). Abstract: Towards Real-world Federated Learning in Medical Image Analysis using Kaapana. In Informatik aktuell (p. 140). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-658-41657-7_31
Mendeley helps you to discover research relevant for your work.