The Radiopaedia Reading Room Podcast 76. AI, fake x-rays and bots, oh my!
Apr 1, 2026
Conversations jump from AI-made deepfakes and fake radiographs to robots and viral bot videos. They debate whether AI could replace multidisciplinary tumor boards and how simple AI tools might help routine radiology work. There are stories about conference antics, a cheeky CAPTCHA April Fools stunt using X-rays, and reflections on learning from mistakes and medical hype.
AI Snips
Chapters
Books
Transcript
Episode notes
Deepfake Radiographs Can Fool Experts
- Deepfakes are now good enough to fool experienced radiologists about 40% of the time on casual review.
- Mount Sinai study showed radiologists only reached ~70–75% accuracy when explicitly asked to detect GPT-generated radiographs, and the fakes often looked 'too clean'.
Require Multi-View DICOMs To Stop Fake Cases
- Strengthen editorial safeguards by requiring multiple views/modalities and DICOM uploads instead of single exported images.
- Radiopaedia changed submission rules and review processes after encountering AI-generated images to reduce fake uploads.
Realistic Images Amplify Health Tech Fraud Risk
- Fake-looking but plausible medical visuals combined with 'fake it till you make it' startup culture raises risk of fraudulent health tech claims.
- Andrew linked the Theranos story to the danger of lifelike generated images being used to sell nonfunctional products.
