AI-Generated 'Deepfake' X-Rays Fool Radiologists and AI Models

Study finds neither medical experts nor language models can easily distinguish artificial X-ray images from authentic ones, raising concerns about fraud and cybersecurity risks.

Mar. 25, 2026 at 1:06am

Got story updates? Submit your updates here. ›

A new study published in the journal Radiology found that neither radiologists nor multimodal large language models (LLMs) were able to easily distinguish artificial intelligence (AI)-generated 'deepfake' X-ray images from authentic ones. The findings highlight the potential risks associated with AI-generated medical images, including fraudulent litigation and cybersecurity vulnerabilities if hackers were to inject synthetic images into hospital networks.

Why it matters

The inability to reliably detect AI-generated X-ray images poses significant risks, as fabricated medical images could be used to support fraudulent legal claims or manipulate patient diagnoses. This also raises cybersecurity concerns, as hackers could potentially infiltrate hospital systems and inject synthetic images to undermine the integrity of digital medical records.

The details

The study involved 17 radiologists from 12 different centers in six countries, with experience ranging from 0 to 40 years. Half of the 264 X-ray images in the study were authentic, while the other half were generated by AI. When radiologists were unaware of the study's purpose, only 41% spontaneously identified the AI-generated images. After being informed that the dataset contained synthetic images, the radiologists' mean accuracy in differentiating the real and synthetic X-rays was 75%. Individual radiologist performance ranged from 58% to 92% in accurately detecting the ChatGPT-generated images, and the accuracy of four multimodal LLMs ranged from 57% to 85%. The study also identified common features of synthetic X-rays, such as overly smooth bones, unnaturally straight spines, and unusually clean fractures.

  • The study was published on March 25, 2026.

The players

Mickael Tordjman, M.D.

A post-doctoral fellow at the Icahn School of Medicine at Mount Sinai in New York and the lead author of the study.

Radiological Society of North America (RSNA)

The organization that publishes the journal Radiology, where the study was published.

ChatGPT

The AI model used to generate some of the synthetic X-ray images in the study.

RoentGen

An open-source generative AI diffusion model developed by Stanford Medicine researchers, which was used to create some of the synthetic chest X-ray images in the study.

GPT-4o, GPT-5, Gemini 2.5 Pro, and Llama 4 Maverick

The multimodal large language models (LLMs) that were evaluated in the study for their ability to detect the AI-generated X-ray images.

Got photos? Submit your photos here. ›

What they’re saying

“Our study demonstrates that these deepfake X-rays are realistic enough to deceive radiologists, the most highly trained medical image specialists, even when they were aware that AI-generated images were present.”

— Mickael Tordjman, M.D., Post-doctoral fellow, Icahn School of Medicine at Mount Sinai

“Deepfake medical images often look too perfect. Bones are overly smooth, spines unnaturally straight, lungs overly symmetrical, blood vessel patterns excessively uniform, and fractures appear unusually clean and consistent, often limited to one side of the bone.”

— Mickael Tordjman, M.D., Post-doctoral fellow, Icahn School of Medicine at Mount Sinai

“We are potentially only seeing the tip of the iceberg. The logical next step in this evolution is AI-generation of synthetic 3D images, such as CT and MRI. Establishing educational datasets and detection tools now is critical.”

— Mickael Tordjman, M.D., Post-doctoral fellow, Icahn School of Medicine at Mount Sinai

What’s next

The study's authors have published a curated deepfake dataset with interactive quizzes for educational purposes to help train radiologists and AI models to better detect synthetic medical images.

The takeaway

This study highlights the growing threat of AI-generated 'deepfake' medical images, which can deceive even the most highly trained radiologists and AI models. Implementing robust digital safeguards, such as watermarking and cryptographic signatures, is crucial to protect the integrity of medical imaging and prevent the potential misuse of these technologies for fraudulent or malicious purposes.