[A Meta-analysis of the application of artificial intelligence in cervical cytopathology diagnosis]

Objective: To systematically evaluate the application of artificial intelligence (AI) in cervical cytopathology diagnosis. Methods: A systematic search was conducted using the keywords ''cervical cancer'' ''cytology'' ''artificial intelligence'' ''sensitivity'' and ''specificity'' (in both English and Chinese) across databases including PubMed, Web of Science, Embase, Cochrane Library, IEEE Xplore, CNKI, Wanfang, VIP Chinese Science and Technology Journals, and SinoMed. The search covered literature from inception until January 1, 2024, on the application of AI in cervical cytopathological diagnosis. Data were extracted using a predefined data extraction form to compile the contingency table data, from which sensitivity, specificity and area under the curve (AUC) were calculated. Results: A total of 1 616 articles were initially retrieved, and 27 articles were finally included in this study according to the inclusion and exclusion criteria. Five researches were conducted on the diagnosis of cytopathology slides, with pooled AUC, sensitivity and specificity of 0.92 (95%CI: 0.89-0.94), 0.91 (95%CI: 0.77-0.97) and 0.84 (95%CI: 0.77-0.90), respectively. About 22 researches were conducted on the diagnosis of cytopathology images (individual cells or cell clusters), with pooled AUC, sensitivity and specificity of 1.00 (95%CI: 0.99-1.00), 0.98 (95%CI: 0.97-0.99) and 0.98 (95%CI: 0.97-0.99), respectively. Conclusion: The application of AI in the field of cervical cytopathology shows certain diagnostic performance and potential clinical application value.