Privacy preserving federated learning for brain tumor segmentation using multi institutional MRI data

Federated brain tumor segmentation using privacy-preserving deep learning has emerged as a promising approach to address the challenges of data scarcity, data heterogeneity, and patient privacy in multi-institutional medical imaging research. Conventional deep learning methods for brain tumor segmentation typically rely on centralized MRI datasets, which are often inaccessible due to institutional regulations, ethical constraints, and patient privacy requirements, thereby limiting the generalizability of models trained on data from a single institution. Federated learning enables collaborative model training across multiple healthcare institutions by allowing each institution to train the model locally while sharing only model parameters or gradients with a central aggregation server, without transferring raw patient MRI data. However, federated learning alone does not inherently guarantee complete privacy, as model updates may still be vulnerable to information leakage. Therefore, additional privacy-preserving mechanisms, such as secure aggregation and encrypted model updates, are incorporated to enhance data confidentiality during collaborative training. This distributed learning paradigm improves the robustness and generalization of segmentation models by learning from diverse patient populations, imaging protocols, and institutional variations while maintaining compliance with data protection regulations. Experimental evaluation on the publicly available BraTS multi-institutional MRI dataset demonstrates that the proposed federated framework achieves competitive segmentation performance compared with centralized learning approaches while preserving patient data privacy. Overall, the proposed privacy-preserving federated deep learning framework provides a scalable, secure, and effective solution for collaborative brain tumor segmentation, supporting future cross-institutional clinical applications and precision medicine.