Global research landscape of large language models in radiology reporting, summarization, and workflow: a bibliometric and knowledge-mapping analysis

Background Large language models (LLMs) are increasingly being explored for radiology-related language tasks, including report generation, report summarization, structured reporting, information extraction, patient-friendly communication, and workflow support. Although the literature has expanded rapidly, the overall research landscape has not been formally mapped through a focused bibliometric analysis. Objective To characterize the global research landscape of LLMs in radiology reporting, summarization, and workflow-oriented applications using bibliometric and knowledge-mapping methods. Methods This bibliometric study analyzed journal articles indexed in OpenAlex and published between January 1, 2020, and April 11, 2026. The search strategy combined terms related to LLMs, radiology, and reporting or workflow tasks, while excluding primarily image-based artificial intelligence applications. After date restriction and document-type filtering, titles and abstracts were screened for relevance. Eligible studies were analyzed for publication growth, citation metrics, leading journals, countries, institutions, open-access status, and thematic distribution. Network visualization of title- and abstract-term co-occurrence and country-level co-authorship was performed using VOSviewer. Results The final dataset included 265 studies. No eligible studies were identified between 2020 and 2022; the field emerged in 2023 with 26 publications, increased to 68 in 2024, and peaked at 134 in 2025, with 37 additional studies indexed by April 11, 2026. Radiology was the most prolific journal, with 18 publications, followed by the Journal of Imaging Informatics in Medicine with 15 and European Radiology with 12. The USA was the leading contributing country, with 89 studies, followed by Germany with 46 and China with 33. Humboldt-Universität zu Berlin was the most productive institution, with 12 publications. The 265 studies accumulated 4577 citations, with a mean of 17.3 citations per article, a median of 3, and an H-index of 32. Open access accounted for 78.1% of the literature. Thematic mapping showed dominant clusters centered on structured reporting, report simplification, impression generation, information extraction, patient-centered communication, and workflow-oriented use of LLMs. Conclusions Research on LLMs in radiology reporting and related workflow tasks has grown rapidly since 2023, with output concentrated in a small number of journals, countries, and institutions. The literature is centered on practical text-based applications rather than autonomous image interpretation. These findings provide a focused map of an emerging field and highlight the need for future work emphasizing external validation, prospective workflow evaluation, privacy-aware deployment, and methodological rigor.