One-sentence headline summary

A study published in Radiology reveals that radiologists and advanced AI models struggle to distinguish between authentic X-rays and synthetic deepfake images created by generative AI tools.

Key points

• Researchers tested 264 X-ray images on 17 radiologists and four multimodal LLMs, including GPT-4o, GPT-5, Gemini 2.5 Pro, and Llama 4 Maverick.
• Radiologists identified AI-generated images with only 41% accuracy when unaware of their presence, improving to 75% once alerted to the potential for fakes.
• Synthetic images often exhibit unnatural characteristics, such as overly smooth bones, symmetrical lungs, and excessively uniform blood vessel patterns.
• The study found no correlation between a radiologist's years of experience and their ability to detect synthetic medical imagery.
• Experts recommend implementing cryptographic signatures and invisible watermarks to verify the authenticity of medical records and prevent fraudulent clinical manipulation.

The inability to reliably verify medical imagery poses significant risks for insurance fraud, legal disputes, and the integrity of digital patient records. As AI-generated 3D scans like MRIs and CTs emerge, developing robust detection tools is essential to maintaining clinical safety and diagnostic accuracy.