LLM Performance Varies for Feedback on Bitewing Radiographic Technique Errors

In a study comparing four LLMs for feedback on bitewing radiograph positioning errors, GPT o3-mini had significantly lower odds of higher scores than Gemini 2.0 Flash, while engineered prompts reduced Grok 3’s performance.

• Positioning error type influenced performance, with significant model-dependent differences across scenario groups.
• No significant interaction between LLM and assessment category (detail, content, clarity, relevance) was observed.

Automated summary

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