人工智能辅助系统在宫颈液基细胞学分析中的应用价值研究  

Study on the Value of Artificial Intelligence-assisted Systems in Cervical Liquid-based Cytology Analysis

作　　者：陈洁 李文生[1] 张巍 CHEN Jie;LI Wensheng;ZHANG Wei(Department of Pathology,Shaanxi Provincial People’s Hospital,Xi’an 710068,China)

机构地区：[1]陕西省人民医院病理科,西安710068

出　　处：《现代检验医学杂志》2023年第5期155-159,共5页Journal of Modern Laboratory Medicine

基　　金：陕西省自然科学基础研究计划(2021JQ-914):BRAF基因突变及VE-1蛋白表达在甲状腺针吸检诊断的研究;陕西省人民医院科技人才支持计划“菁英人才”项目(2021JY-50):BRAF基因在甲状腺乳头状癌亚型中的表达差异研究。

摘　　要：目的探讨人工智能(artificial intelligence,AI)辅助系统在宫颈薄层液基细胞学检查分析中的应用价值。方法收集2020年1~8月在陕西省人民医院行宫颈细胞学检查的620例液基病例,其中经膜式薄层液基细胞学(thinprep cytology test,TCT)制片技术制片426例,经沉降法液基细胞制片技术(liquid-based cytology technology,LCT)制片194例。同时由病理科中级医师及AI辅助系统阅片,以高级医师审核结果为最终结果进行对比分析。结果中级医师在TCT组及LCT组中的分类判读结果与高级医师审核结果相比较,其差异均无统计学意义(χ^(2)=0.594,1.014,P=0.898,0.602)。AI辅助系统在TCT组及LCT组中的分类判读结果均与高级医师审核结果不一致,其差异具有统计学意义(χ^(2)=104.267,26.349,均P<0.001)。AI辅助系统在TCT,LCT两种制片方式中对阳性片的检测灵敏度分别为100%和83.33%,高于中级医师(86.36%,66.67%),其差异无统计学意义(χ^(2)=3.220,0.444,P=0.233,1.000);两种制片方式AI辅助系统判读特异度为65.10%和84.57%,均低于中级医师(99.26%,100%),差异具有统计学意义(χ^(2)=160.931,31.424,均P<0.001);准确度为66.90%和84.54%,均低于中级医师(98.59%,98.97%),差异具有统计学意义(χ^(2)=149.831,24.829,均P<0.001)。结论AI辅助系统在TCT及LCT两种制片方式中灵敏度与中级医师相似,有利于阳性病例的筛出,但准确度及特异度较低,需要结合高级医师审核结果以提高准确度。Objective To investigate the value of artificial intelligence(AI)assisted system in the application of cervical thin layer cytology.Methods 620 liquid-based cases of cervical cytology were collected from Shaanxi Provincial People’s Hospital,of which 426 cases were produced by thinprep cytologic test(TCT)and 194 cases were produced by liquid-based cytology technology(LCT).All smears were interpreted by intermediate pathologists and AI-assisted systems,and the results were reviewed by senior doctors for comparison and analysis as thefinal results.Results The difference between the intermediate pathologist’s classification interpretation results in the TCT and LCT groups compared to the senior pathologist’s results was not statistically significant(χ^(2)=0.594,1.014,P=0.898,0.602).The difference between the classification interpretation results of the AI-assisted system in the TCT and LCT groups compared to the senior pathologist’s results was statistically significant(χ^(2)=104.267,26.349,all P<0.001).The sensitivity of the AI-assisted system in detecting positive slices in both TCT and LCT production methods was 100%and 83.33%for pathologists(86.36%and 66.67%),respectively,and the difference was not statistically significant(χ^(2)=3.220,0.444,P=0.233,1.000).The two the specificity of AI-assisted system interpretation was 65.10%and 84.57%,both lower than that of pathologists(99.26%,100%),and the difference was statistically significant(χ^(2)=160.931,31.424,all P<0.001).The accuracy was 66.90%and 84.54%,both lower than that of pathologists(98.59%,98.97%),and the differences were statistically significant(χ^(2)=149.831,24.829,all P<0.001).Conclusion The AI-assisted system was similar in sensitivity to the intermediate pathologist in both TCT and LCT methods of production,facilitating the screening of positive cases,but the accuracy and specificity were low and need to be combined with senior pathologist review to improve accuracy.

关 键 词：宫颈薄层液基细胞学 人工智能辅助系统 沉降法液基细胞制片技术 膜式薄层液基细胞学制片技术 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] R446.8[自动化与计算机技术—控制科学与工程]