LAUSR

Clinical utility, accuracy, and efficiency are the keys to costeffective medical procedures. The Papanicolaou test, while highly useful, is plagued by problems in accuracy and work intensity. This combination of need and difficulty led to intense investigation in the 1950s and 1960s into computerization to enhance the diagnostic accuracy of gynecologic cytology and to enhance workflow efficiency. Since these initial investigations, computerbased technology has advanced such that morphologic features in cells can be translated into data points for algorithmic analysis and subsequent identification of squamous abnormalities, leading to the popularization of technologybased platforms such as Hologic’s ThinPrep Imaging System. Continued research in the past decade has further enhanced image analysis algorithms to improve performance of artificial intelligence (AI) in cytology screening. The benefits of integrating AI into clinical practice include its capacity to increase detection sensitivity and decrease subjectivity and diagnostic gray zones by providing quantitative data and analysis in real time during the morphologic assessment of specimens. Despite these documented improvements, AI has not been widely implemented into clinical practice. The most significant obstacles to its popularization include higher upfront costs of investment, disruption of historical workflows, a perceived negative impact on job security, and the regulatory burden (and hence cost) of bringing such products to market. However, as technology and AI algorithms further develop, more avenues are being explored to minimize these and other concerns. In the corresponding article, Tang et al beautifully highlight a practical example of how innovative AI applications can be implemented to leverage the advantages of technology in resourcelimited regions while bypassing the need for large, expensive hardware or an overhaul of workflow. Addressing the lack of experienced practitioners for optimal manual review of cervical cytology specimens and for training, the authors introduce a novel AI microscope that combines the conventional features of microscopy with augmented reality and an AIempowered computer unit. The results yield a realtime visualization of the AI analysis in the optical path of evaluated slides for immediate user feedback. This application pivots from the foundation of previous AI systems, which have largely focused on whole slide imaging with subsequent postevaluation guidance to selected fields/cells of interest or whole slide classification/scoring. Instead, the authors have created an opportunity for realtime AI assistance. The utility of the authors’ AI algorithm was demonstrated in the 2round reader study of the publication, in which 4 masked cytopathologists looked at a total of 486 Papanicolaou specimens twice: the first time without AI assistance and the second time with AI assistance 15 days later. Use of the AI microscope enhanced the sensitivity of detection of squamous abnormalities designated as lowgrade squamous intraepithelial lesion and higher and also increased diagnostic agreement among the cytopathologists. These promising results showcase how AI not only can enhance workflow efficiency but also can enhance