人工智能辅助诊断在宫颈液基薄层细胞学中的应用

Application of artificial intelligence-assisted diagnosis for cervical liquid-based thin-layer cytology

目的探讨宫颈液基薄层细胞学TBS报告人工智能辅助诊断系统在宫颈癌筛查中的应用价值。方法收集2020年7至9月间南方医科大学附属南方医院、广州华银医学检验中心、深圳市宝安人民医院(集团)及长沙远安生物科技有限公司共16 317例宫颈液基薄层细胞涂片临床样本及相关资料。利用南方医科大学与锟元方青医疗科技有限公司联合开发的基于深度学习卷积神经网络的宫颈液基薄层细胞学TBS报告人工智能辅助诊断系统,对所有临床样本进行人工智能辅助诊断。以2014版的TBS为评价标准,分析人工智能辅助诊断系统及细胞病理医师运用人工智能辅助诊断系统阅片的灵敏度、特异度及准确度,并比较两种阅片方式所用的时间。结果人工智能辅助诊断系统在不同制片方式、不同胞质染色及不同仪器扫描下预测宫颈上皮内病变的灵敏度为92.90%,其他病变(包括>45岁妇女子宫内膜细胞及感染性病变)灵敏度为83.55%,阴性样本特异度为87.02%;而细胞病理医师运用人工智能辅助诊断系统分别为99.34%,97.79%及99.10%,且比人工镜下阅片节省约6倍的阅片时间。结论宫颈液基薄层细胞学TBS报告人工智能辅助诊断系统具有高灵敏度、高特异度及强泛化性等优势,细胞病理医师运用人工智能辅助诊断系统阅片能显著提高阅片的准确率和工作效率。

Objective To explore the application value of artificial intelligence-assisted diagnosis system for TBS report in cervical cancer screening.Methods A total of 16317 clinical samples and related data of cervical liquid-based thin-layer cell smears,which were obtained from July 2020 to September 2020,were collected from Southern Hospital,Guangzhou Huayin Medical Inspection Center,Shenzhen Bao′an People′s Hospital(Group)and Changsha Yuan′an Biotechnology Co.,Ltd.The TBS report artificial intelligence-assisted diagnosis system of cervical liquid-based thin-layer cytology jointly developed by Southern Medical University and Guangzhou F.Q.PATHOTECH Co.,Ltd.based on deep learning convolution neural network was used to diagnose all clinical samples.The sensitivity,specificity and accuracy of both artificial intelligence-assisted diagnosis system and cytologists using artificial intelligence-assisted diagnosis system were analyzed based on the evaluation standard(2014 TBS).The time spent by the two methods was also compared.Results The sensitivity of artificial intelligence-assisted diagnosis system in predicting cervical intraepithelial lesions and other lesions(including endometrial cells detected in women over 45 years old and infectious lesions)under different production methods,different cytoplasmic staining and different scanning instruments was 92.90%and 83.55%respectively,and the specificity of negative samples was 87.02%,while that of cytologists using artificial intelligence-assisted diagnosis system was 99.34%,97.79%and 99.10%,respectively.Moreover,cytologists using artificial intelligence-assisted diagnosis system could save about 6 times of reading time than manual.Conclusions Artificial intelligence-assisted diagnosis system for TBS report of cervical liquid-based thin-layer cytology has the advantages of high sensitivity,high specificity and strong generalization.Cytologists can significantly improve the accuracy and work efficiency of reading smears by using artificial intelligence-assisted diagnosis system.