Study Reveals Challenges in Detecting AI-Generated X-Rays

A new study published in the journal Radiology reveals that radiologists and advanced artificial intelligence models alike struggle to distinguish realistic X-ray images generated by AI from authentic clinical scans, raising alarms about potential misuse in healthcare.

The research, led by Mickael Tordjman, a postdoctoral fellow at the Icahn School of Medicine at Mount Sinai in New York, involved 17 radiologists from 12 medical centers across six countries: the United States, France, Germany, the United Arab Emirates, the United Kingdom, and Turkey. Participants ranged from trainees to experts with up to 40 years of experience, spanning subspecialties such as musculoskeletal, thoracic, and body imaging. Conducted between April and August 2025, the retrospective analysis tested detection abilities using two balanced datasets totaling 264 images, half real and half synthetic.

In the first dataset, 77 authentic radiographs from clinical databases were paired with 77 AI-generated versions created by OpenAI's GPT-4o model using simple text prompts specifying anatomy, conditions, and noise levels. These covered chest, extremities, and spine views. The second dataset featured 55 real...