Prompt engineering enables open-source large language models to match proprietary models in diagnostic accuracy for annotation of radiology reports

This study investigates whether open-source large language models (LLMs) can achieve diagnostic accuracy comparable to proprietary models in annotating trauma radiology reports in a low-resource language. The findings demonstrate that with effective prompt engineering, small open-source LLMs can accurately identify clinical findings, achieving high accuracy rates that rival those of proprietary models, thus providing a viable and privacy-conscious alternative for clinical applications.