Clinical Report: A Federated Learning Approach for Privacy-Enhanced Translation of CBCT to Synthetic CT in Head and Neck Imaging

Overview

This study presents a federated learning approach to synthesize synthetic CT from cone-beam computed tomography in head and neck imaging while preserving data privacy. The model demonstrated effective generalization across multiple centers, achieving comparable performance metrics on external validation datasets.

Background

Cone-beam computed tomography (CBCT) is increasingly utilized in image-guided radiotherapy (IGRT), but it suffers from limitations such as noise and artifacts that affect Hounsfield unit accuracy. These issues hinder the reliable estimation of electron density for dose calculations. The development of synthetic CT (sCT) from CBCT images aims to improve dosimetric fidelity, making this research critical for enhancing treatment precision in radiotherapy.

Data Highlights

The federated model achieved the following performance metrics: mean absolute error (MAE), structural similarity index (SSIM), and peak signal-to-noise ratio (PSNR) on both training and external validation datasets, confirming robust cross-center generalization.

Key Findings

• The federated learning model effectively synthesized synthetic CT from CBCT images while maintaining data privacy.
• Mean absolute error (MAE) and other performance metrics indicated strong generalization across different medical centers.
• Visual analysis revealed that registration errors influenced the obtained metrics.
• The approach utilized a combination of FedAvg and FedProx aggregation strategies for model training.
• Data from three European medical centers were used in the study, showcasing the feasibility of multi-center collaboration without data sharing.

Clinical Implications

The findings suggest that federated learning can facilitate the development of generalizable models for synthetic CT generation in head and neck imaging, potentially improving treatment planning and delivery in radiotherapy. This approach may also address data privacy concerns in multi-center studies.

Conclusion

The study demonstrates the technical feasibility of using federated learning for synthesizing synthetic CT from CBCT while preserving data privacy, paving the way for collaborative advancements in radiotherapy imaging.

Related Resources & Content

1. Automating C-arm Alignment for Standard Imaging Views in Orthopedic Procedures, Springer, 2020 -- Automating C-arm Alignment for Standard Imaging Views in Orthopedic Procedures
2. An Adaptive Learning Approach for Real-Time Modification of C-arm Cone-beam CT Source Paths to Minimize Artifacts, Springer, 2020 -- An Adaptive Learning Approach for Real-Time Modification of C-arm Cone-beam CT Source Paths to Minimize Artifacts
3. An Effective Sampling Method for Accurate Distribution Simulation in Federated Learning, Springer, 2025 -- An Effective Sampling Method for Accurate Distribution Simulation in Federated Learning
4. Quality and Safety Considerations for Adaptive Radiation Therapy: An ASTRO White Paper, PubMed, 2023 -- Quality and Safety Considerations for Adaptive Radiation Therapy: An ASTRO White Paper
5. Dose calculation accuracy of clinical radiotherapy plans using next generation cone beam computed tomography imaging technology, PubMed, 2025 -- Dose calculation accuracy of clinical radiotherapy plans using next generation cone beam computed tomography imaging technology
6. CT-Based Mandible Segmentation for Virtual Surgical Planning Utilizing an Enhanced Two-Stage Convolutional Neural Network
7. AAPM Reports - Quality assurance for image-guided radiation therapy utilizing CT-based technologies
8. AAPM Reports - Use of image registration and fusion algorithms and techniques in radiotherapy
9. Cone beam CT dose optimisation: A review and expert consensus by the 2022 ESTRO Physics Workshop IGRT working group
10. Quality and Safety Considerations for Adaptive Radiation Therapy: An ASTRO White Paper - PubMed
11. Dose calculation accuracy of clinical radiotherapy plans using next generation cone beam computed tomography imaging technology - PubMed
12. Feasibility of HyperSight CBCT for adaptive radiation therapy: A phantom benchmark study of dose calculation accuracy and delivery verification on the Halcyon - PubMed
13. Modeling dose uncertainty in cone-beam computed tomography: Predictive approach for deep learning-based synthetic computed tomography generation - ScienceDirect
14. An automated cone-beam CT dosimetric assessment pipeline for adaptive head and neck radiotherapy | Journal of Radiotherapy in Practice | Cambridge Core
15. Dosimetric Advantages of Online Adaptive Radiation Therapy for Head and Neck Squamous Cell Carcinoma: Results From a Prospective Registry Study - ScienceDirect
16. Contour uncertainty assessment for MD-omitted daily adaptive online head and neck radiotherapy - PubMed
17. Generating synthetic computed tomography for radiotherapy: SynthRAD2025 challenge report
18. A joint learning framework for multisite CBCT-to-CT translation using a hybrid CNN-transformer synthesizer and a registration network - PubMed