Related Experiment Video
Updated: May 6, 2026
08:25
Radiation Planning Assistant - A Streamlined, Fully Automated Radiotherapy Treatment Planning System
Published on: April 11, 2018
15.3K
Federated prediction for scalable and privacy-preserved knowledge-based planning in radiotherapy
Jingyun Chen1, David Horowitz1,2, Yading Yuan1,3
1Department of Radiation Oncology, Columbia University Irving Medical Center, New York, NY, USA.
Arxiv
|June 5, 2025
Summary
Federated learning platforms like FedKBP+ enhance radiation therapy planning by overcoming data challenges. FedKBP+ shows effectiveness and efficiency in predictive tasks, improving model generalizability across institutions.
Area of Science:
- Medical Physics
- Artificial Intelligence
- Radiotherapy
Background:
- Deep learning in radiation therapy planning faces challenges in model generalizability due to data scarcity and heterogeneity.
- Sharing data across institutions is difficult due to privacy and technical obstacles.
Purpose of the Study:
- To develop FedKBP+, a federated learning (FL) platform for predictive tasks in radiotherapy treatment planning.
- To address the dilemma of data sharing while improving model performance.
Main Methods:
- Implemented a unified communication stack using Google Remote Procedure Call (gRPC).
- Developed both centralized and fully decentralized FL models with Peer-to-Peer communication.
- Evaluated FedKBP+ on three predictive tasks using the scale-attention network (SA-Net) model.
Main Results:
- FedKBP+ with FedAvg outperformed local training and matched centrally trained models for 3D dose prediction (OpenKBP Challenge).
- FedKBP+ surpassed NVFlare in accuracy and efficiency for brain tumor segmentation (BraTS challenge).
- A novel decentralized FL algorithm in FedKBP+ demonstrated robustness against site failures in organ segmentation (PanSeg dataset).
Conclusions:
- FedKBP+ is a highly effective, efficient, and robust federated learning platform for radiation therapy.
- The platform shows significant potential for improving clinical adoption of AI in radiotherapy planning.
