Saieg, Mauro

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.

BACKGROUNDThe Papanicolaou Society of Cytopathology has recently published a new proposal for pancreaticobiliary cytology terminology. For the current study, the authors applied this new classification for pancreaticobiliary cytology to a set of previously diagnosed endoscopic ultrasound (EUS)‐guided fine‐needle aspiration (EUS‐FNA) samples and correlated the results with those from the corresponding available final surgical specimens.METHODSPancreatic FNA specimens were retrospectively collected from a 12‐month period at the Mayo Clinic in Jacksonville, Florida. Cases with cytohistologic disagreement were reviewed to assess the reasons for discrepancies. The sensitivity and specificity of EUS‐FNA specimens were assessed using the Pearson chi‐square test.RESULTSOne hundred fifty‐five pancreatic FNA specimens were retrieved. The median patient age was 70 years (range, 25‐93 years). Eleven specimens previously classified as negative and 3 previously classified as non‐diagnostic were reclassified as neoplastic:other, as well as all specimens primarily classified as atypical or positive for neoplasm and half the specimens primarily classified as suspicious. All positive specimens remained within their same categories in the revised classification. Sixteen patients (10.3%) had surgical resection specimens available, and complete or partial agreement with FNA results was achieved in 13 of those surgical specimens (81.2%). Reasons for discrepancy comprised sampling errors in 5 specimens and an interpretation error in 1 specimen. Overall, sensitivity was 66.7%, specificity was 66.7%, the positive predictive value was 88.9%, and the negative predictive value was 33.3%.CONCLUSIONSThe application of the new proposed terminology for pancreaticobiliary cytology had a greater impact among specimens that were previously classified as atypical and suspicious. EUS‐FNA of the pancreas is a highly accurate method, and its use, allied with the new proposed terminology, may ultimately contribute to better outcomes. Cancer (Cancer Cytopathol) 2015;123:488–94. © 2015 American Cancer Society.