AI tools could rewrite radiology reports in plainer language, with patients rating them almost twice as easy to understand, a new review suggests.
Researchers analysed 38 studies published between 2022 and 2025, covering more than 12,000 radiology reports simplified using AI systems such as ChatGPT.
Patients, the public and clinicians assessed the rewritten reports.
Across X-ray, CT and MRI reports, the reading level fell from “university level” to one closer to that of an 11 to 13-year-old.
The review, led by researchers at the University of Sheffield, suggests AI explanations could sit alongside scan reports as access expands through services such as the NHS App and wider medical record transparency policies.
Dr Samer Alabed is senior clinical research fellow at the University of Sheffield and honorary consultant cardio radiologist at Sheffield Teaching Hospitals NHS Foundation Trust.
Alabed said: “The fundamental issue with these reports is they’re not written with patients in mind.
“They are often filled with technical jargon and abbreviations that can easily be misunderstood, leading to unnecessary anxiety, false reassurance and confusion.
“Patients with lower health literacy or English as a second language are particularly disadvantaged.
“Clinicians frequently have to use valuable appointment time explaining report terminology instead of focusing on care and treatment. Even small time savings per patient could add up to significant benefits across the NHS.”
Clinicians said most AI-simplified reports were accurate and complete, but around 1 per cent contained errors, including an incorrect diagnosis, suggesting the approach still needs oversight.
None of the 38 studies were carried out in the UK or in NHS settings, a gap the research team said it is now working to address.
Alabed said: “This research has highlighted several key priorities.
“The most important is the need for real world testing in NHS clinical workflows to properly assess safety, efficiency, and patient outcomes.
“This includes human-oversight models, where clinicians review and approve AI-generated explanations before they are shared with patients.
“Our long-term goal is not to replace clinicians, but to support clearer, kinder, and more equitable communication in healthcare.”
