Answers are grounded in this paper and cite their sources.
Objective To map large language model (LLM) applications in uroradiology and assess their deployment maturity, clinical risk, and reproducibility. Materials and Methods PRISMA-ScR-compliant scoping review of PubMed/MEDLINE, Scopus and Web of Science (to 17 July 2026). Two independent reviewers classified eligible LLM studies on genitourinary imaging reports by task type and intended use, along a five-level deployment maturity continuum, with indicators of transparency, reproducibility, and hallucination reporting. Results Twenty-nine studies (2023-2026) were included, addressing the prostate (n=19, e.g., PI-RADS scoring), kidney (n=7, e.g., Bosniak classification), and bladder (n=3, e.g., VI-RADS staging), among 27 with an organ-specific focus. Twenty-three (79.3%) reached Maturity Level 2 (retrospective validation on real clinical data); the remaining six (20.7%) used synthetic scenarios (Level 1). None achieved external validation, prospective testing, or clinical deployment. By task type, most studies (23/29, 79.3%) generated a new score, diagnosis, or recommendation rather than only extracting an already-documented value; all 29 (100%) were retrospective research or registry analyses. GPT-4 variants predominated (75.9%). A reproducibility paradox emerged: 96.6% disclosed prompts and 100% defined reference standards, yet only one study released evaluation code and only seven reported inference parameters. Hallucinations were discussed in 44.8%, with inconsistent definitions. Conclusion A maturity-risk mismatch characterises the field: most studies generate new clinical outputs rather than extracting existing data, yet all were evaluated only retrospectively and lack the reproducibility infrastructure for safe clinician-facing deployment. Advancing requires multicentre external validation (Level 3), then monitored prospective pilots (Level 4), with LLM-specific reporting standards and governance.
No takes yet. Share an insight, caveat, or question.
Salvaggio et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: