Radiologists and leading AI models struggle to distinguish AI-generated deepfake X-ray images from authentic radiographs, according to a recent Radiology journal study.
Key Details
- 1Study involved 264 X-ray images, half real and half generated by AI models, including ChatGPT-4o and RoentGen.
- 217 radiologists from six countries participated, with accuracy in identifying deepfake images ranging from 58% to 92%.
- 3Four leading multimodal LLMs (GPT-4o, GPT-5, Gemini 2.5 Pro, Llama 4 Maverick) had detection accuracies between 52% and 89%.
- 4No correlation was found between years of radiology experience and accuracy; musculoskeletal specialists performed better than others.
- 5Common AI-deepfake X-ray features included overly smooth bones, symmetric lungs, and unnaturally straight spines.
- 6Proposed safeguards include invisible watermarks and cryptographic image signatures.
Why It Matters
These findings expose high-stakes risks for clinical practice, legal proceedings, and hospital cybersecurity as synthetic images become increasingly authentic. Urgent adoption of digital safeguards and specialized training is needed to protect the integrity of medical imaging.
Source
EurekAlert
Related News
•EurekAlert
AI Clinical Reasoning in Diagnostics and Digital Fatigue in Healthcare
Recent JMIR features explore large language models in clinical diagnostics and digital fatigue among healthcare professionals.
•EurekAlert
KAIST, MIT, Microsoft Develop Efficient AI Image Upsampling for Robotics
KAIST, MIT, and Microsoft have created 'Upsample Anything,' a training-free AI method to restore high-resolution visual data from compressed images with up to 16x improved GPU memory efficiency.
•EurekAlert
AI Method Protects Sensitive Data in ECGs While Retaining Clinical Value
Researchers developed an AI model (PP-VAE) that safeguards sensitive personal data in electrocardiograms without sacrificing clinical utility.