Researchers at the University of Nottingham Ningbo China (UNNC) have developed an AI-based medical imaging technology that could reduce patients’ exposure to radiation during PET scans, while maintaining image quality.
The research, led by UNNC PhD student Meiyuan Wen and her supervisor, Professor Xiangjian He, Chair Professor of Computer Science, has been published in npj Digital Medicine, a Nature Portfolio medical journal.
PET/CT scans are widely used in cancer diagnosis and follow-up. While PET provides information about the body’s metabolic activity, CT is traditionally used to correct signal loss in PET images caused by tissues inside the body. However, CT also exposes patients to additional radiation and can cause image-matching issues due to respiratory movement.
The UNNC-led team developed an AI model, CrossPET-Adapt, which can perform PET attenuation and scatter correction without the need for CT. The model was tested on 1,539 cases across 11 cohorts, covering four radioactive tracers, four scanners and four medical centres. Its corrected images and key indicators of lesions were found to be highly consistent with those produced using conventional CT-based methods.
A key advantage of the technology is its ability to adapt quickly to new hospitals, scanners and radioactive tracers using only a small amount of local data. In the fastest cases, the model required just one to five samples and around nine minutes to adapt, significantly reducing the data and computing resources needed for deployment in new clinical settings.
For Wen, the project also highlights the value of interdisciplinary research. Her work brought together computer science, medical imaging and clinical expertise, with collaboration from researchers at the Shenzhen Institute of Advanced Technology (SIAT), Chinese Academy of Sciences, and Shanghai Ruijin Hospital.
“Publication is not the end of our exploration,” says Wen. “This summer, I have expanded my research into low-dose PET imaging, exploring ways to reduce the amount of radioactive tracer used or shorten scanning times.”
The team plans to validate the technology across more hospitals, countries and scanning devices, with the aim of bringing AI-powered, lower-dose medical imaging closer to real-world clinical applications.
Published on 16 September 2026