利用人工智能图像识别系统诊断子宫内膜细胞病理学的有效性研究  

Effectiveness of the artificial intelligence image recognition system in diagnosing endometrial cytopathology

作　　者：安静 尹盼月 王斌 史桂芝 钟德星[4] 王建六[5] 李奇灵[1] AN Jing;YIN Panyue;WANG Bin;SHI Guizhi;ZHONG Dexing;WANG Jianliu;LI Qiling(Department of Obstetrics and Gynecology,The First Affiliated Hospital of Xi’an Jiaotong University,Xi’an 710061;Department of Obstetrics and Gynecology,Xi’an Daxing Hospital,Xi’an 710016;Savaid Medical School,University of Chinese Academy of Sciences,Beijing 101408;School of Automation Science and Engineering,Xi’an Jiaotong University,Xi’an 710049;Department of Obstetrics and Gynecology,Peking University People’s Hospital,Beijing 100044,China)

机构地区：[1]西安交通大学第一附属医院妇产科,陕西西安710061 [2]西安大兴医院妇产科,陕西西安710016 [3]中国科学院大学存济医学院,北京101408 [4]西安交通大学自动化科学与工程学院,陕西西安710049 [5]北京大学人民医院妇产科,北京100044

出　　处：《西安交通大学学报（医学版）》2024年第2期343-347,共5页Journal of Xi’an Jiaotong University（Medical Sciences）

基　　金：陕西省重点研发计划-重点项目(No.2017ZDXM-SF-068);西安交通大学第一附属医院临床研究重点项目(No.XJTU1AF-CRF-2019-002);陕西省技术创新引导专项(No.2019QYPY-138)。

摘　　要：目的探讨基于人工智能(artificial intelligence,AI)的图像识别系统对子宫内膜细胞团块良恶性诊断的有效性。方法选取2021年8月至2023年2月西安交通大学第一附属医院和西安大兴医院的子宫内膜细胞学标本,以组织病理学为金标准,对比分析AI图像识别系统(AI诊断)和专业病理医师人工诊断(人工诊断)子宫内膜细胞团块良恶性的灵敏度、特异度、阳性预测值、阴性预测值、准确率和诊断所需时间。结果纳入分析的126例患者中,AI诊断与组织学诊断的总体符合率为92.1%(116/126),与组织学病理结果高度一致(Kappa=0.841);人工诊断和组织学诊断的总体符合率为94.4%(119/126),与组织学病理结果高度一致(Kappa=0.889)。AI诊断与人工诊断两种方法差异无统计学意义(χ^(2)=0.568,P=0.451)。AI诊断的灵敏度、特异度、阳性预测值和阴性预测值分别为91.8%、92.3%、91.8%和92.3%。126张细胞学切片,人工诊断每张切片所需6.67 min;AI诊断每张切片所需5.00 min。结论AI图像识别系统具有较高的诊断准确性、灵敏度和特异度,与专业病理医师人工诊断水平相当,在诊断子宫内膜细胞团块良恶性方面具有应用价值。Objective To explore the effectiveness of an image recognition system based on artificial intelligence(AI) in diagnosing benign and malignant endometrial cell clumps.Methods We selected endometrial cytological specimens from The First Affiliated Hospital of Xi'an Jiaotong University and Xi'an Daxing Hospital from August 2021 to February 2023;histopathology was used as the gold standard.We compared and analyzed the sensitivity,specificity,positive predictive value,negative predictive value,accuracy and diagnostic time of AI image recognition system(AI diagnosis) and professional pathologists' manual diagnosis(manual diagnosis) of benign and malignant endometrial cell clumps.Results Among the 126 patients included in the analysis,the overall coincidence rate of AI diagnosis and histological diagnosis was 92.1%(116/126),which was highly consistent with histopathological results(Kappa=0.841).The overall coincidence rate of manual diagnosis and histological diagnosis was 94.4%(119/126),which was highly consistent with histopathological results(Kappa=0.889).There was no statistically significant difference between AI diagnosis and manual diagnosis methods(χ^(2)=0.568,P=0.451).The sensitivity,specificity,positive predictive value,and negative predictive value of AI diagnosis were 91.8%,92.3%,91.8%,and 92.3%,respectively.There were 126 cytology sections,each of which required 6.67 minutes for manual diagnosis and 5.00 minutes for AI diagnosis.Conclusion The AI image recognition system has high diagnostic accuracy,sensitivity and specificity,which is equivalent to the manual diagnosis level of professional pathologists.Therefore,this system has application value in the diagnosis of benign and malignant endometrial cell clumps.

关 键 词：子宫内膜癌 人工智能(AI) 细胞学 诊断 有效性 

分 类 号：R737.33[医药卫生—肿瘤]