Federated Learning for Organs-At-Risk Segmentation in Prostate Cancer: A Multi-Institutional CT Study

Overview & Synopsis

Radiotherapy (RT)-based treatments offer a precise therapeutic option for prostate cancer. Accurate delineation of the prostate and surrounding organs-at-risk (OARs) is essential to maximize radiation dose delivery to the tumor while minimizing exposure to nearby healthy tissues. Deep learning (DL)-based methods have emerged as a preferred approach for generating fast and accurate segmentations. However, their effectiveness depends heavily on access to large, diverse, and well-annotated datasets. In highly regulated clinical environments, managing and sharing such sensitive medical data is often expensive, time-consuming, and constrained by privacy regulations.To address these challenges, we propose a federated learning (FL)-based approach for the segmentation of OARs in computed tomography (CT) images used for prostate cancer RT planning. We evaluate and compare FL against conventional centralized learning, and further explore different FL training strategies. Our findings suggest that FL can achieve segmentation performance comparable to centralized methods, while preserving patient data privacy across institutional boundaries.