Application of large language models in medical diagnosis: A bibliometric review

Background The integration of Large Language Models into medical diagnosis represents an emerging field with the potential to support diagnostic workflows across diverse clinical settings. However, the trends and evolutionary trajectory of LLM-assisted diagnostic research remain insufficiently understood. Objective This bibliometric review aims to map the global research landscape, identify key research clusters, and analyze the development trajectory of LLM technologies in medical diagnosis, with an emphasis on descriptive synthesis rather than formal evaluation. Methods A bibliometric analysis was conducted on relevant publications retrieved from the Web of Science Core Collection, covering the period from Q1 2023 to Q1 2025. The extracted data were processed and visualized using Excel, ArcGIS, VOSviewer, CiteSpace, and Pajek. The analyses included publication trends, influential authors and institutions, collaboration networks, and research cluster mapping. Results A total of 650 publications were included in the analysis. Research output increased markedly from Q1 2023 onward, rising from 2 publications to 148 by Q1 2025, corresponding to an average quarterly growth rate of 71.25%. The United States (273 publications), China (135 publications), and Germany (65 publications) emerged as the leading contributing countries. The three most productive institutions were all based in the United States: Harvard University (26 publications), Stanford University (26 publications), and the Icahn School of Medicine at Mount Sinai (20 publications). Keyword co-occurrence analysis identified 10 core clusters, with a modularity Q value of 0.8231 and a silhouette S value of 0.9412, indicating a highly coherent clustering structure and strong internal consistency. Conclusion The development of LLM technologies has substantially influenced the research landscape of medical diagnostics. As this field continues to evolve, it is crucial to refine model performance, integrate multimodal data, and address ethical considerations. Future research should focus on optimizing LLMs for specific clinical applications and evaluating their implementation in real-world healthcare settings.