Meta-analysis: Generative AI Less Accurate than Physicians in Diagnostic Radiology

In 48 studies, physicians outperformed generative AI on diagnostic radiology tasks by 13.0 percentage points (95% CI 0.9-25.2; P=.038). AI accuracy: 42.9% (free-text) and 58.1% (choice tasks).

• Text-only input was associated with higher AI accuracy than image-only (difference -25.9 percentage points; P=.001) or text-and-image input.

Automated summary

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