Comparative Analysis of Large Language Models Performance in Appropriate Diagnostic Imaging Modality Selection

This study aimed to compare the performance of seven large language models (LLMs) in selecting appropriate diagnostic imaging modalities based on ACR guidelines. Using 50 clinical vignettes, the authors assessed the models' accuracy, clinical reasoning, and citation quality, finding that while all models provided reasonable imaging recommendations, there was significant variability in citation validity, particularly with Google's Gemini models. The results highlight the importance of aligning clinical reasoning with reliable sources before implementing LLMs in clinical practice.