基于云技术和人工智能的自动化细胞学诊断平台在703103名女性宫颈癌筛查中的应用研究  

Application of artificial intelligence and cloud-based diagnostic platform in population-based screening of cervical cancer:A study of 703103 cases

作　　者：李艳辉 周星 徐琴 蔡俐琼 梁铭霖 闵洁 汪宏波[1] LI Yanhui;ZHOU Xing;XU Qin;CAI Liqiong;LIANG Minglin;MIN Jie;WANG Hongbo(Department of Obstetrics and Gynecology,Union Hospital,Tongji Medical College,Hua Zhong University of Science and Technology,Wuhan,Hubei 430022,China)

机构地区：[1]华中科技大学同济医学院附属协和医院妇产科,武汉430022

出　　处：《中国妇产科临床杂志》2022年第4期364-368,共5页Chinese Journal of Clinical Obstetrics and Gynecology

基　　金：国家科技攻关计划(2018YFC0114605)。

摘　　要：目的探讨一种基于云技术和人工智能(artificial intelligence,AI)的自动化细胞学诊断平台应用于人群宫颈癌筛查的有效性和准确率。方法通过构建一种基于云技术和AI的自动化细胞学诊断平台;并于2018年1—12月,将该平台应用于来自湖北省83个县(市)、年龄20~72岁的703103名女性的宫颈癌筛查;对云技术+AI自动化诊断平台筛查出的“上皮细胞异常”的阳性样本及随机抽样10%阴性样本,由有经验的细胞病理学医生复核。此外,经AI自动化平台筛查出的高度鳞状上皮内病变、低度鳞状上皮内病变合并高危型人乳头状病毒感染者均建议行阴道镜检查+宫颈活检确诊;分析该自动化细胞学诊断平台用于宫颈癌筛查的准确性和有效性。结果AI自动化细胞学诊断平台在人群宫颈癌筛查中的成功率为96.76%(680344/703103)。AI自动化诊断平台与人工阅片相比,在未见上皮内瘤变及恶性细胞、非典型鳞状细胞-意义不明、高度提示存在高级别的宫颈病变、低度鳞状上皮内病变、高度鳞状上皮内病变的诊断一致率分别为99.10%(63943/64521)、87.49%(21326/24376)、84.15%(7069/8400)、85.31%(1185/1389)和94.20%(877/931)。以宫颈活检组织病理学结果作为金标准,AI自动化诊断平台的敏感性和特异性分别为99.18%和44.42%。而阴性预测值(NPV)和阳性预测值(PPV)分别为41.27%和99.28%。结论基于云技术和AI的自动化细胞学诊断平台用于宫颈癌筛查,具有极高的效率和准确性,值得在人群普查中推广应用。Objective To explore the effectiveness and accuracy of an automated cytological diagnosis platform based on cloud technology and artificial intelligence applied to large-scale cervical cancer screening.Methods An automated cytological diagnostic system based on cloud technology and AI was constructed and applied to 703103women aged 20~72 years in 83 counties(cities)of Hubei Province for cervical cancer screening in 2018.All positive cases of“epithelial abnormality”and 10%of randomly selected cases with normal cytological results diagnosed by the AI diagnostic platform were reviewed by experienced cytopathologists.In addition,colposcopy plus cervical biopsy were recommended for high-grade cervical intraepithelial lesions and low-grade cervical intraepithelial lesions with high-risk HPV.The accuracy and efficiency of AI diagnostic platform for cervical cancer screening were analyzed.Results The success rate of AI automated diagnosis platform in this study was 96.76%(680344/703103).The consistent diagnostic rates between AI automatic diagnosis platform and manual reading,for the normal cytological results,atypical squamous cells of undetermined significance(ASC-US),atypical squamous cells-cannot exclude HSIL(ASC-H),lowgrade squamous intraepithelial lesion(LSIL)and high-grade squamous intraepithelial lesion(HSIL),were 99.10%(63943/64521),87.49%(21326/24376),84.15%(7069/8400),85.31%(1185/1389)and 94.20%(877/931)respectively.The sensitivity and specificity of AI platform were 99.18%and 44.42%,respectively,when compared with cervical biopsy histopathological results.Negative predictive value(NPV)and positive predictive value(PPV)were 41.27%and 99.28%respectively.Conclusion The cloud-based and AI automated cytological diagnosis platform constructed in this study has high efficiency and accuracy in cervical cancer screening,and is worthy of application in large-scale population screening.

关 键 词：人工智能 云技术 宫颈癌筛查 大规模人群筛查 

分 类 号：R737.33[医药卫生—肿瘤]