MIMIC-RD：大语言模型能否在真实临床环境中进行罕见病的鉴别诊断？

This study introduces MIMIC-RD, a novel benchmark for rare disease differential diagnosis that addresses critical limitations in existing approaches. While rare diseases affect 1 in 10 Americans, their diagnosis remains challenging. The researchers developed a methodology that directly maps clinical text entities to Orphanet, overcoming the limitations of both idealized case studies and ICD code-based approaches. The process involved an initial LLM-based mining phase followed by validation from four medical annotators. Through evaluation of various models on a dataset of 145 patients, the study revealed significant performance gaps in current state-of-the-art LLMs for rare disease differential diagnosis. This work highlights the substantial distance between existing AI capabilities and real clinical needs, while providing valuable insights and directions for future improvements in rare disease diagnosis.