人工智能辅助诊断系统在液基细胞学宫颈腺上皮病变筛查中的应用价值  

Value of an artificial intelligence-assisted diagnostic system in screening of glandular epithelial lesions based on liquid cervical smear

作　　者：刘颖[1] 纪晓坤 吴娟[1] 张艳[1] 赵银环 吴家宁[1] 王蕊[1] 郭晓 杜芸[1] LIU Ying;JI Xiaokun;WU Juan;ZHANG Yan;ZHAO Yinhuan;WU Jianing;WANG Rui;GUO Xiao;DU Yun(Cancer Detection Center,The Fourth Hospital of Hebei Medical University,Shijiazhuang 050011,Hebei,China)

机构地区：[1]河北医科大学第四医院癌检中心,河北石家庄050011

出　　处：《癌变．畸变．突变》2024年第5期391-394,共4页Carcinogenesis,Teratogenesis & Mutagenesis

基　　金：河北省重点研发计划项目-民生科技专项(20377723D)。

摘　　要：目的:探讨人工智能(AI)辅助系统在宫颈腺上皮病变细胞学诊断中的应用价值。方法:收集诊断为非典型腺细胞(AGC)的液基细胞学薄层宫颈涂片143例,阴性涂片631例,所有涂片均进行AI辅助阅片和中级医师阅片,以主任医师复核的诊断结果为金标准,进行对比分析,并统计特异性、敏感性等指标。结果:宫颈涂片腺上皮细胞病变的检测中,AI辅助阅片系统的阳性率为15.7%、特异性为99.8%,而中级医师阅片的阳性率为18.3%、特异性为99.2%,两组间差异无统计学意义(P>0.05)。AI辅助系统的准确率为97.0%,敏感性为84.6%,中级医师阅片的准确率为99.2%,敏感性为99.3%,两组间差异具有统计学意义(P<0.05)。AI辅助阅片的ROC曲线下面积(AUC)为0.922,低于中级医师阅片的0.993,差异有统计学意义(P<0.05)。此外,AI辅助阅片与中级医师阅片的诊断一致率达到99%,相应的Kappa值为0.888,表明两种阅片方法基本一致。结论:AI辅助系统在宫颈腺上皮病变的诊断中展现出较高的特异性,然而其敏感性相对较低,存在一定的漏诊可能性。尽管AI辅助阅片的准确率尚未达到中级医师的水平,但其仍展现出重要的诊断潜力。未来的研究和发展应着重于提升AI辅助阅片系统的敏感性,并通过临床验证来确保其在实际应用中的可靠性和安全性。OBJECTIVE:To investigate the value of an artificial intelligence(AI)assistant system in cytological diagnosis of cervical glandular epithelial lesions.METHODS:A total of 143 liquid-based thin layer cervical cytological smears diagnosed as atypical adenocyte(AGC)and 631 negative smears were collected.AI assistant system and intermediate pathologist diagnosis were performed on all smears.The diagnostic results of senior physicians were used as the gold standard for comparative analysis,and specificity,sensitivity and other indicators were statistically analyzed.RESULTS:The positive rate and specificity of AI-assisted diagnosis system were 15.7%and 99.8%,while those of intermediate pathologist diagnosis were 18.3%and 99.2%respectively.There was no significant difference between the two groups(P>0.05).The accuracy and sensitivity of AI-assisted system were 97.0%and 84.6%,and the accuracy and sensitivity of intermediate pathologist diagnosis were 99.2%and 99.3%,and the difference between the two groups was statistically significant(P<0.05).The area under ROC curve(AUC)of AI-assisted diagnosis was 0.922,which was lower than that of intermediate pathologist diagnosis(0.993),and the difference was statistically significant(P<0.05).In addition,the diagnostic agreement rate between AI-assisted diagnosis and intermediate pathologist diagnosis reached 99%,and the corresponding Kappa value was 0.888,indicating that the two diagnostic methods were basically consistent.CONCLUSION:The AI-assisted system has high specificity in the diagnosis of cervical glandular epithelial lesions,but its sensitivity is low,and there is a certain risk of missed diagnosis.Although the accuracy of AI-assisted diagnosis is not as good as that of intermediate pathologist diagnosis,it still had a high diagnostic value.Therefore,the AI assistant system should be further improved and optimized in the diagnostic application of cervical glandular epithelial cells.

关 键 词：人工智能 液基细胞学 辅助阅片 宫颈腺上皮病变 

分 类 号：R737.33[医药卫生—肿瘤]