Fake X-ray images created by artificial intelligence to resemble true results from human patients can fool not only experienced radiologists but also the AI tools themselves, according to a study that illustrates the potential for manipulation by bad actors.
Seventeen radiologists from 12 hospitals in six countries reviewed 264 X-ray images, half of which had been generated by the AI tools ChatGPT or RoentGen.
When radiologist readers were unaware of the study's true purpose, only 41% spontaneously identified AI-generated images, according to a report published in Radiology.
After being informed that the dataset contained synthetic images, the radiologists' mean accuracy in differentiating the real and synthetic X-rays rose to 75%.
Having deepfake X-rays realistic enough to deceive radiologists "creates a high-stakes vulnerability for fraudulent litigation if, for example, a fabricated fracture could be indistinguishable from a real one," study leader Dr. Mickael Tordjman of the Icahn School of Medicine at Mount Sinai in New York said in a statement.
"There is also a significant cybersecurity risk if hackers were to gain access to a hospital's network and inject synthetic images to manipulate patient diagnoses or cause widespread clinical chaos by undermining the fundamental reliability of the digital medical record," Tordjman said.
The accuracy of four large language models - GPT-4o (OpenAI), GPT-5 (OpenAI), Gemini 2.5 Pro (Google), and Llama 4 Maverick (Meta Platforms) - at detecting the fake images ranged from 57% to 85%.
Even ChatGPT-4o, the model that created the deepfakes, failed to detect all of them, though it identified more than the other LLMs, the researchers reported.
Potential digital safeguards are needed to help distinguish real and fake images and prevent tampering such as use of invisible watermarks that embed ownership, researchers said.
"We are potentially only seeing the tip of the iceberg," said Tordjman of the eventual possibility of fake CT and MRI scans. "Establishing educational datasets and detection tools now is critical."
No comments:
Post a Comment