Privacy-Preserving AI Tool Synthesizes CT for Radiation Therapy Planning

Radiation therapy planning depends on accurate medical imaging to calculate dose, visualize organs, and position patients during treatment. Hospitals need large imaging datasets to train robust artificial intelligence models, but sensitive patient information is difficult to share because of privacy rules and organizational barriers. These limits can slow development of tools that support clinical workflows. Researchers have now developed privacy-friendly, secure, and more sustainable AI techniques for medical imaging in radiation therapy.

Researchers at Karlsruhe Institute of Technology (KIT; Karlsruhe, Germany) investigated federated learning techniques for radiation therapy applications. Federated learning allows a shared artificial intelligence model to be trained across decentralized sites while patient data remain at the institutions where they were collected. Instead of transferring sensitive health information, participating hospitals exchange only model updates, allowing knowledge from multiple facilities to be combined without centralizing patient data.

The work focused on synthesizing computed tomography (CT) images from other imaging modalities used in radiation therapy planning. The researchers developed and tested methods for generating synthetic CT images of the brain and the head and neck region. These approaches are relevant because CT is required for dose calculation, while MRI provides high-contrast visualization of organs and cone-beam CT supports patient positioning during treatment.

The researchers found that federated models could be trained across multiple hospitals without exchanging sensitive patient data. They also examined whether information from individual patients could be reconstructed from a trained artificial intelligence model and reported that relatively minor adaptations were sufficient to strengthen federated systems against this type of data-security attack.

The findings were reported across several publications in Computers in Biology and Medicine in 2025, Frontiers in Digital Health in 2026, the Journal of Medical Systems in 2026, and Sustainable Computing: Informatics and Systems in 2026. The work included FedSynthCT-Brain, a federated learning framework for multi-institutional brain MRI-to-CT synthesis, as well as a privacy-preserving framework for cone-beam CT-to-synthetic CT translation in head and neck imaging. In the sustainability-focused work, training time and carbon dioxide emissions were reduced by 9% to 23% without affecting synthetic image quality.

“Data protection alone is not enough. If federated systems are to be used in medical practice in the future, we have to take potential security risks into account right from the start,” said Ciro Benito Raggio, a researcher in Professor Maria Francesca Spadea's Biomedical Imaging for Modeling and Simulation research group at Karlsruhe Institute of Technology's Institute of Biomedical Engineering. 

“Depending on the model architecture, we were able to reduce the time required for the training and CO₂ emissions by 9% to 23% without affecting the quality of the synthetic images. Smaller hospitals in particular could benefit from such optimizations, as they often don't have a lot of computing resources,” added Raggio.

Related Links
Karlsruhe Institute of Technology