Vision-language models match large language models for labeling incidental breast findings on CT

A retrospective study found vision-language models (VLMs) performed on par with LLMs for labeling incidental breast findings on CT reports, achieving similar accuracy without substantial performance gaps (exact figures not reported in source).

• Presented at SIIM 2025; study compared VLMs against LLMs for extracting and labeling breast findings from CT reports.
• Design and sample size not detailed in the source snippet (exact n, validation methodology unreported).
• Limitation: Single-conference abstract without peer-reviewed publication; external generalizability remains unproven.

Automated summary

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