Related Experiment Video
Updated: Jul 13, 2026
Dynamic Lung Tumor Tracking for Stereotactic Ablative Body Radiation Therapy
Published on: June 7, 2015
Center-specific Federated Learning for Radiation Pneumonitis: A Cross-Center Adaptive Alternating Framework
Meng Yan1, Zhixiang Wang2, Liyu Ning3
1Department of Radiation Oncology, Zhejiang Cancer Hospital, Hangzhou Institute of Medicine (HIM), Chinese Academy of Sciences, Zhejiang Key Laboratory of Particle Radiotherapy Equipment, Hangzhou, China; Department of Radiation Oncology, Key Laboratory of Cancer Prevention and Therapy, Tianjin Medical University Cancer Institute and Hospital, National Clinical Research Center for Cancer, Tianjin's Clinical Research Center for Cancer, Tianjin, China; Department of Radiation Oncology (Maastro), GROW Research Institute for Oncology and Reproduction, Maastricht University Medical Centre+, Maastricht, The Netherlands.
A new federated learning approach (FCAAM) improves radiation pneumonitis (RP) prediction across institutions. This privacy-preserving method enhances clinical decision-making for safer radiotherapy by overcoming data heterogeneity challenges.
Area of Science:
- Medical Physics
- Artificial Intelligence in Medicine
- Radiotherapy Oncology
Background:
- Accurate prediction of symptomatic radiation pneumonitis (RP) is crucial for radiotherapy planning.
- Deep learning model generalization is limited by restricted multi-center data access.
- Standard federated learning (FL) algorithms struggle with heterogeneous clinical data.
Purpose of the Study:
- To evaluate the clinical feasibility of a center-specific federated learning approach for RP prediction.
- To assess the performance of the Federated Cross-Center Adaptive Alternating Model (FCAAM) in a multi-center setting.
- To address the limitations of standard FL in handling data heterogeneity and privacy concerns.
Main Methods:
- Evaluation of FCAAM, a framework decoupling global representations from center-specific adaptations.
- Utilized dynamic weighting for data heterogeneity and differential privacy for security.
- Tested FCAAM on 1,238 patients across four diverse datasets for RP prediction using CT and dose images.
Main Results:
- FCAAM demonstrated superior cross-center performance and robustness compared to single-center and FedAvg models.
- Achieved a stable AUC of 0.71-0.77 across test sets, outperforming baseline models.
- FCAAM's performance was comparable to centralized models and showed improved sensitivity on smaller datasets.
Conclusions:
- FCAAM offers a privacy-preserving, robust, and interpretable solution for multi-center RP prediction.
- This center-specific strategy can enhance clinical decision-making and reduce performance disparities.
- The approach supports safer, personalized radiotherapy through improved multi-center collaboration.
