Objective: To systematically summarize and evaluate the current application status and research progress of artificial intelligence (AI) in cervical cancer screening in China. Methods: Literature related to the application of AI in cervical cancer screening in China was searched in PubMed, Embase, Cochrane Library, IEEE, China National Knowledge Infrastructure (CNKI), and Wanfang Database using the keywords"cervical cancer","artificial intelligence","screening","machine learning","deep learning","neural network","uterine cervical neoplasms,"uterine cervical tumor","diagnosis", and"China". The search was limited to studies published in Chinese and English. As of July 2025, a total of 35 eligible articles were included. Basic information from the included studies was extracted and summarized. In addition, the National Medical Products Administration (NMPA) official website was searched using the term"cervix"to identify approved AI-assisted cervical cancer screening products. Results: A total of 21 AI-assisted cervical cancer screening technologies were identified, including 17 technologies for primary screening, mainly AI-assisted cytology, and 4 technologies for colposcopic diagnosis. For AI-assisted cytology, the sensitivity ranged from 67.5% to 100.0% and the specificity ranged from 9.9% to 99.8% in hospital-based populations, with the overall accuracy of some technologies exceeding 90%. In community-based screening populations, the sensitivity ranged from 83.0% to 100.0% and the specificity ranged from 74.2% to 99.9%. Most studies suggested that AI could improve the diagnostic performance of pathologists to some extent, shorten the average slide-reading time, and enhance overall screening efficiency. A total of 24 AI-assisted cervical cancer screening products have been approved by the NMPA, all of which are AI-assisted cytology technologies, and corresponding studies were identified for 8 of these products. For AI-assisted colposcopic diagnosis used as a standalone screening modality, the sensitivity and specificity for identifying cervical intraepithelial neoplasia grade 2 (CIN2) or worse ranged from 43.6% to 95.5% and from 51.8% to 93.9%, respectively; for cervical intraepithelial neoplasia grade 3 (CIN3) or worse, the sensitivity ranged from 35.1% to 97.5% and the specificity ranged from 56.6% to 87.2%. In the physician-assisting mode, the sensitivity increased to 95.1%-97.5%, with improvements in interobserver consistency and diagnostic accuracy among less experienced colposcopists. Conclusions: AI has shown promising potential in cervical cancer screening in China. However, more scientific evidence is needed to determine whether it can be effectively integrated into the existing cervical cancer prevention and control system in China.
目的： 系统总结和评价人工智能（AI）技术在我国宫颈癌筛查中的应用现状与研究进展。 方法： 以“宫颈癌”“人工智能”“筛查”以及“artificial intelligence”“machine learning”“deep learning”“neural network”“cervical cancer”“uterine cervical neoplasms”“uterine cervical tumor”“screening”“diagnosis”“China”为关键词，检索PubMed、Embase、Cochrane、IEEE、中国知网和万方数据库中关于AI技术应用于宫颈癌筛查的文献，语种限定为中文和英文。截至2025年7月，共纳入35篇有效文献，对纳入文献的基本信息进行摘录和整理分析。在国家药品监督管理局（NMPA）官方网站以“宫颈”为检索词，检查AI辅助宫颈癌筛查产品获批情况。 结果： 共涉及21项AI辅助宫颈癌筛查技术，其中AI辅助宫颈癌初筛技术17项，以AI辅助细胞学检查技术为主；AI辅助阴道镜诊断技术4项。AI辅助细胞学检查技术在医院就诊人群中的灵敏度为67.5%~100.0%，特异度为9.9%~99.8%，部分技术总体准确率超过90%；在社区筛查人群中的灵敏度为83.0%~100.0%，特异度为74.2%~99.9%。研究普遍表明，AI可在一定程度上提升病理医师的判读效能，有效缩短平均阅片时间，提升整体筛查效率。NMPA目前共批准24款AI辅助宫颈癌筛查产品，均为AI辅助细胞学检查技术，其中有8款检索到对应研究。AI辅助阴道镜诊断技术在独立筛查模式下识别宫颈上皮内瘤变Ⅱ级（CIN2）及以上病变的灵敏度为43.6%~95.5%，特异度为51.8%~93.9%；识别宫颈上皮内瘤变Ⅲ级（CIN3）及以上的灵敏度为35.1%~97.5%，特异度为56.6%~87.2%。在辅助医师模式下，灵敏度提升至95.1%~97.5%，可改善低年资阴道镜医师的判读一致性和准确率。 结论： AI在我国宫颈癌筛查中展现出一定潜力，但其能否嵌入我国现有宫颈癌防控体系仍需更多科学证据。.