Radiologists Struggle to Detect AI-Generated X-Rays
A new award-winning study reveals that medical experts and AI models struggle to identify synthetic medical imaging.
Updated on Oct. 20, 2026 in Artificial Intelligence
Live Poll
Do you trust the ability of current medical professionals to detect AI-generated clinical images?
The Radiological Society of North America has awarded the 2026 Alexander R. Margulis Award to researchers for their study on detecting AI-generated x-rays. The findings suggest that both radiologists and multimodal LLMs face significant challenges in identifying synthetic images.
Why it matters
The research highlights the growing difficulty of distinguishing between authentic diagnostic data and synthetic images produced by advanced AI models. As these technologies become more integrated into healthcare, the inability of experts and LLMs to consistently detect synthetic content poses a critical challenge for medical integrity.
The study involved 17 radiologists across 12 centers in 6 countries, with four multimodal LLMs failing to detect every synthetic image. Mean accuracy for identifying AI-generated content reached 74.8% for GPT-4o and 70% for the RoentGen model.
The players
Radiological Society of North America
This professional organization is the primary body dedicated to advancing the field of radiology through research, education, and practice standards.
Mickael Tordjman
He is a researcher who co-led the study examining the efficacy of radiological identification of synthetic medical imaging.
Murat Yuce
He is a researcher who served as a co-lead on the investigation regarding AI-generated content within medical diagnostic tools.
The details
Led by researchers Mickael Tordjman and Murat Yuce, the study evaluated the ability of 17 radiologists and four multimodal LLMs to differentiate between authentic and synthetic radiographs. Remarkably, professional experience provided no advantage in identifying the AI-generated content, with only seven radiologists noticing the synthetic nature of the images spontaneously.
Timeline
October 20, 2026: The research article documenting the award win was published.
November 29, 2026: The award presentation will take place at the RSNA 2026 conference in Chicago.
The Tech Race
This research reflects the rapid evolution of generative AI, which is now capable of producing medical imagery that challenges even the most experienced professionals. It positions the medical community in an ongoing race to develop verification standards as legacy diagnostic systems confront the proliferation of synthetic data.
For healthcare providers and patients, this research signals a future where diagnostic image authentication may become a necessary part of the medical review process. While no immediate clinical changes are required, it highlights the need for clinicians to remain vigilant when interpreting images that could potentially be AI-generated.
The takeaway
The study demonstrates that human expertise and current AI models are not yet foolproof in distinguishing synthetic radiographs from the real thing. It serves as a reminder for the medical field to prioritize the development of robust provenance tools as diagnostic imaging evolves.
Further reading
Learn more about the latest innovations and challenges in Artificial Intelligence.
Source note: This article includes information reported by AuntMinnie.
Live Poll
Do you trust the ability of current medical professionals to detect AI-generated clinical images?
