The increasing adoption of AI-driven medical imaging is constrained by strict privacy requirements and institutional data-governance policies. Federated Learning (FL) offers a practical solution by enabling collaborative model training without sharing raw patient data. This work presents a privacy-preserving federated segmentation framework for breast imaging using the Kaapana platform. Through the nnunet-federated-workflow, multiple sites jointly train a 3D nnU-Net model while keeping all imaging data on-premise; only encrypted model updates are exchanged with a central aggregator. After training, the global model is deployed locally at each site via the nnunet-predict workflow. The framework is evaluated on a controlled synthetic phantom dataset across two federated environments. The global model achieved near-perfect performance for Circle and Square, with median Dice scores of 0.983 and 0.986, median ASD values below 0.05 mm, and median Hausdorff distances around 1.4 mm. Triangle segmentation remained substantially more challenging, with a median Dice of 0.38 and a median Hausdorff distance of 11 mm, indicating strong class-specific variation. Overall, the results show that Kaapana’s federated learning components deliver a scalable, reproducible, and privacy-preserving approach for distributed medical image segmentation, capable of producing highly accurate global models while maintaining strict data locality.