Cytopathology in focus
Elena Enbom, MD, PhD
Xiaoying Liu, MD
May 2026—Digital pathology has evolved substantially over the past decade. What started primarily as a platform for image storage now supports educational initiatives, remote signout, consultation, and image analysis deployment. This digital transformation is now impacting cytopathology in practical ways. For example, the cytopathology board examination incorporates whole slide digital images of smears, liquid-based preparations, cell blocks, and small core biopsies.
With that digital foundation in place, it is not surprising that artificial intelligence and machine learning are beginning to move into the clinical space, potentially influencing how cytology specimens are processed, reviewed, and interpreted. The expected benefits are straightforward and include improved diagnostic performance and greater efficiency and reproducibility. In this article, we review the current landscape of AI applications in cytopathology.
Gynecologic cytology: early adopter. Gynecologic cytology continues to lead digital and AI adoption. Early automation (imager) systems such as BD FocalPoint1 and the Hologic ThinPrep imager paved the way for current deep-learning–based platforms that classify cells and flag high-risk fields of view.
Hologic’s Genius Digital Diagnostics System represents the next generation of AI-powered cytology, integrating advanced image analysis with a digital review environment to improve detection accuracy and streamline workflow. The system enables pathologists to navigate digitized slides more efficiently while leveraging AI to highlight areas of concern.2 This model, which was trained on a very large data set, is built on cell-level classification, with a particular focus on squamous categories. Its underlying framework is based on the morphologic criteria established through successive iterations of The Bethesda System for Reporting Cervical Cytology, which serve as the foundation for instrument training and model development. A prior issue of the CAP TODAY Cytopathology in Focus section highlighted a single-institution experience with validation, implementation, and use of the Hologic system, and the institution’s findings were published last fall.3 Additional abstracts have been presented at national pathology meetings, and future publications are forthcoming on validation, quality, and further emerging applications from this platform.
Urine cytology: promising next step. Urine cytology represents another opportunity for AI deployment. Defined morphologic criteria derived from the Paris System for urine cytology has led to several studies4 and products in the field. Currently, AIxMed’s AIxURO system is the only commercial AI product for urinary cytology. Furthermore, additional studies using AI-based image analysis have focused on automating the Paris System for diagnosis, providing further support for its potential in urine cytology.5,6 Early data from AIxURO and others demonstrate improved reproducibility and efficiency, but further validation is needed. As larger, clinical data sets are produced, urine cytology could become one of the next areas where AI-driven screening and surveillance are applied.
Thyroid cytology: early evidence of AI utility. Thyroid cytology has benefited from the transition to liquid-based preparations, which facilitate image capture and are more amenable to algorithm development. Several studies suggest that AI can assist in applying the Bethesda System for Reporting Thyroid Cytopathology in a more standardized and automated fashion. Early work has also demonstrated the ability to predict diagnostic categories and, in some cases, limited molecular alterations directly from cytologic images, including approaches that leverage publicly available large language models such as ChatGPT in combination with relatively simple, custom imaging setups.7,8
Other cytology areas: building momentum. Rapid onsite evaluation (ROSE) remains a central component of cytopathology practice, allowing for real-time adequacy assessment and triage during fine-needle aspiration procedures, and it represents a natural entry point for AI integration. Early prototypes demonstrate that AI can detect diagnostic clusters and estimate adequacy in real time,9 and recent society abstracts are beginning to describe platforms that provide immediate feedback to cytopathologists and proceduralists. At the same time, pulmonary cytology research is expanding in this space, with AI-assisted models showing the ability to identify diagnostic cell clusters, perform digital stain conversion, and potentially predict key molecular alterations such as EGFR, KRAS, and ALK. In the pancreatobiliary tract, similar approaches have been applied to grading atypia, assisting with adequacy assessment, and predicting treatment response. These data remain preliminary and are largely derived from small data sets, but larger, multi-institutional studies are likely forthcoming.