LLM-Radiologist Collaboration Boosts Dermatologic Ultrasound Diagnostic Accuracy to 83%

In a retrospective analysis of 202 dermatologic US images, adding a radiologist's report to ChatGPT boosted diagnostic accuracy to 83.2%, vs 55.4% (radiologist-only) and 26.2% (ChatGPT-only) (p<0.001).

• Evaluated 202 dermatologic ultrasound images from a public dataset under three reporting conditions: radiologist-only, ChatGPT-only, and ChatGPT with radiologist's report.
• Combined condition achieved superior next-step suggestion accuracy: 77.7% vs 38.6% (radiologist-only) and 59.9% (ChatGPT-only), p<0.001.
• Limited to a single public dataset of dermatologic US; real-world clinical generalizability not demonstrated.

RadPigeon summaries are original and for information only. They are not clinical advice.