-
1
Cone-beam computed tomography (CBCT) is widely used in image-guided radiotherapy but has limitations like noise and artifacts affecting Hounsfield unit values.
-
2
The proposed federated learning approach enables the synthesis of synthetic CT from CBCT while maintaining data privacy across multiple medical centers.
-
3
A conditional generative adversarial network was trained using data from three European centers, demonstrating effective generalization across different imaging protocols.
-
4
Performance metrics for the federated model included mean absolute error, structural similarity index, and peak signal-to-noise ratio, confirming robust cross-center generalization.
-
5
The study highlights the technical feasibility of federated learning for CBCT-to-sCT synthesis, allowing collaborative model development without data sharing.
