人工智能在胃镜活检标本病理诊断中的应用  

Application of artificial intelligence in pathological diagnosis of gastroscopic biopsies

作　　者：孙洁[1] 赵晓婷 张君杰 孙虓[1] 王士娜[1] 吴非[1] 王翠芳[1] 栾岚[1] Sun Jie;Zhao Xiaoting;Zhang Junjie;Sun Xiao;Wang Shina;Wu Fei;Wang Cuifang;Luan Lan(Pathology Department,Central Hospital of Shenyang Medical College,Shenyang 110024,China;Pathology Department,Shenyang Di An Medical Laboratory,Shenyang 110027,China)

机构地区：[1]沈阳医学院附属中心医院病理科,沈阳110024 [2]沈阳迪安医学检验所病理科,沈阳110027

出　　处：《中国组织化学与细胞化学杂志》2022年第5期481-487,共7页Chinese Journal of Histochemistry and Cytochemistry

基　　金：沈阳市科学技术计划公共卫生研发专项基金(21-172-9-18)。

摘　　要：目的评估人工智能图像分析软件的分类准确性并阐明在临床工作中使用自动化扫描系统的要求。方法收集2016年1月至2017年12月之间的胃镜活检组织切片共623例,其中良性288例,恶性(以腺癌为主)335例。至少两名经验丰富的病理专家依据修订维也纳分类评估每张切片的病理诊断,确定每个病例最终一致的诊断;比较了人类病理学家与人工智能图像分析软件的两级(良性或恶性)分类结果。结果应用人工智能图像分析软件分类结果的准确率为0.9871,其中,将正常误判成癌症为0.001,将癌症误判成正常为0.012。结论人工智能自动筛查系统可有效用于不需要人工复查的标本,尤其是阴性标本,从而极大地减轻外科病理专家的负担。Objective To evaluate the classification accuracy of artificial intelligence image analysis software and clarify the requirements for using automated scanning systems in clinical work.Methods A total of 623 gastroscopy biopsies were collected from January 2016 to December 2017,among which 288 were benign and 335 were malignant(mainly adenocarcinoma).At least two experienced pathologists evaluated each slide based on the revised Vienna Classification to determine the final diagnosis with consistency.We compared the two-tier(benign or malignant)classification between human pathologists and artificial intelligence image analysis software.Results The accuracy of classification was 0.9871;the possibility of misjudging normal as cancer was 0.001,whilst the possibility of misjudging cancer as normal was 0.012.Conclusion The artificial intelligence automatic screening system can be effectively used for specimens that do not need to be re-examined mannually,especially negative specimens,thus greatly reducing the burden of surgical pathology experts.

关 键 词：胃活检 人工智能 自动图像分析 

分 类 号：R365[医药卫生—病理学] TP39[医药卫生—基础医学]