Project PI:
- PD Dr. Florian Putz
Department of Radiation Oncology, Universitätsklinikum Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg
florian.putz@uk-erlangen.de - Dr. Stefan Knippen
Department of Radiation Oncology, Helios Clinics of Schwerin-University Campus of MSH Medical School Hamburg, Schwerin, Germany
Stefan.Knippen@helios-gesundheit.de
Project Design:
Multi-center collaborative project utilizing a peer-to-peer federated learning approacxh for the development of deep learning auto-contouring models in radiation oncology. Each participating center trains the model on their local data and shares only the model weights, ensuring data privacy without the need for centralized infrastructure.
Project objectives:
- Collaboratively develop and evaluate advanced AI models for tumor auto-contouring and target volume definition.
- Ensure equal participation and knowledge sharing among centers.
- Maintain patient data privacy by sharing only model parameters, not patient data.
- Eliminate the need for centralized infrastructure while promoting collaborative model improvement.
- Facilitate publication and serve as an introductory project for centers new to AI research.
Project status:
- First application brain metastases auto-segmentation (Collaboration with DEGRO AG Stereotaxie)
- Successful testing of the peer-to-peer federated learning approach between Erlangen and Zurich, with a first application case brain metastasis auto-detection and auto-contouring.
- Two-center collaboration results published in Radiotherapy and Oncology, Volume 198, 110419.
- Currently role-out for multi-center implementation first including Schwerin and then subsequent centers.
- Later expansion to additional application cases, such as anatomically defined target volumens.
- If you like to participate please reach out to florian.putz@uk-erlangen.de or stefan.kippen@helios-gesundheit.de