Lin, Oscar

Abstract Large language models (LLMs) and vision‐language models represent a fundamentally different category of artificial intelligence (AI) compared to prior image analysis approaches in digital pathology, which have largely been based on convolutional neural network architectures. This review from the American Society of Cytopathology Clinical Practice Committee examines the current evidence for LLM and vision‐language model applications in cytopathology, including structured reporting, diagnostic assistance, quality control, education, and workflow integration. The distinction between applications with preliminary evidence and those that remain hypothetical is described. A detailed assessment of the challenges that must be addressed before clinical deployment, including hallucination risk, limited explainability, bias, data privacy, validation gaps, and infrastructure barriers is discussed. A review of the regulatory landscape in the United States and European Union as it applies to AI‐enabled software as a medical device is provided. Recommendations addressing cytopathology‐specific benchmarks, multi‐institutional validation, transparent governance, and incremental deployment beginning with low‐risk applications are suggested. In the current environment, LLMs have the potential to augment cytopathology practice, but responsible adoption requires rigorous validation and sustained collaboration among cytopathologists, AI researchers, and regulatory bodies.

BackgroundFine needle aspiration (FNA) is commonly used for the preoperative evaluation of salivary gland tumors. Tumor grade is a key factor influencing clinical management of salivary gland carcinomas (SGCs). To assess the ability to grade nonbasaloid SGCs in FNA specimens, an international panel of cytopathologists convened to review and score SGC cases.MethodsThe study cohort included 61 cases of primary SGC from the pathology archives of 3 tertiary medical centers. Cases from 2005 to 2016 were selected, scanned, and digitized. Nineteen cytopathologists blinded to the histologic diagnosis reviewed the digitized cytology slides and graded them as low, high, or indeterminate. The panelists' results were then compared to the tumor grades based on histopathologic examination of the corresponding resection specimens.ResultsAll but 2 of the 19 (89.5%) expert panelists review more than 20 salivary gland FNAs per year; 16 (84.2%) of the panelists work at academic medical centers, and 13 (68.4%) have more than 10 years' experience. Participants had an overall accuracy of 89.4% in the grading of SGC cases, with 90.2% and 88.3% for low‐ and high‐grade SGC, respectively. Acinic cell carcinoma and mucoepidermoid carcinoma had the highest degree of accuracy, while epithelial‐myoepithelial carcinoma and salivary duct carcinoma had the lowest degree of accuracy. As expected, the intermediate‐grade SGC cases showed the greatest variability (high‐grade, 42.1%; low‐grade, 37.5%, indeterminate, 20.4%).ConclusionThis study confirms the high accuracy of cytomorphologic grading of primary SGC by FNA as low‐ or high‐grade. However, caution should be exercised when a grade cannot be confidently assigned.