人工智能辅助技术在进展期胃癌原发灶不同区域HER-2评估及判断预后中的应用  

Application of artificial intelligence-assisted technology in assessment of HER-2 expression in different regions of primary lesions of advanced gastric cancer and prognostic estimation

作　　者：刘洋[1] 郭子阳 刘月平[2] 王心然 王子涵 杨家轩 丁平安 郑涛 田园[1] 郭洪海[1] 檀碧波[1] 范立侨[1] 李勇[1] 赵群[1] LIU Yang;GUO Ziyang;LIU Yueping;WANG Xinran;WANG Zihan;YANG Jiaxuan;DING Pingan;ZHENG Tao;TIAN Yuan;GUO Honghai;TAN Bibo;FAN Liqiao;LI Yong;ZHAO Qun(The Third Department of General Surgery,the Fourth Hospital of Hebei Medical University,Shijiazhuang 050011,China;Department of Pathology,the Fourth Hospital of Hebei Medical University,Shijiazhuang 050011,China)

机构地区：[1]河北医科大学第四医院外三科,河北石家庄050011 [2]河北医科大学第四医院病理科,河北石家庄050011

出　　处：《中国普通外科杂志》2023年第4期566-574,共9页China Journal of General Surgery

基　　金：政府资助临床医学优秀人才培养基金资助项目(冀财社[2019]139号);河北省胃癌精准诊断与综合治疗重点实验室基金资助项目(SZX2022014)。

摘　　要：背景与目的:胃癌中人类表皮生长因子受体2(HER-2)表达存在较高的异质性,全面地评估患者的HER-2状态有利于发现抗HER-2治疗的潜在受益者。人工智能(AI)辅助显微镜可以通过大量读片以综合判断HER-2的状态,并减少人为评估的视觉误差。本研究旨在探讨AI辅助显微镜在胃癌原发灶多区域HER-2评估中的实用性和可行性。方法:对264例进展期胃癌患者术后标本同一区域HER-2免疫组化染色切片,分别使用普通光学显微镜视觉评估、AI辅助显微镜两种方法进行评估,评价AI辅助显微镜在胃癌HER-2评估中的准确性;并使用AI辅助显微镜对上述胃癌患者原发灶的其他区域的HER-2表达进行评估,将两个区域HER-2评分较高者作为最终判读结果,并进一步分析HER-2过表达与进展期胃癌患者临床病理特征及术后生存的关系。结果:视觉评估、AI辅助显微镜与金标准三者之间HER-2评分总体差异无统计学意义(P>0.05),但AI辅助显微镜相较视觉评估与金标准具有更高的一致性(κ=0.86 vs.κ=0.81,P<0.05)。使用AI辅助显微镜对胃癌原发灶两个不同区域的HER-2评估结果显示,两个区域HER-2评分不一致者为55例,不一致率为20.8%。其中区域1 HER-2过表达率为29.9%,区域2 HER-2过表达率为31.0%,综合评估两个区域取较高评分的HER-2过表达率为35.2%。综合评估两个区域HER-2表达得出的HER-2过表达率高于单独判读一个区域,但差异无统计学意义(P>0.05);进一步分析发现,胃癌原发灶HER-2异质性表达与肿瘤分化程度及Lauren分型明显有关(均P<0.05)。生存分析显示,HER-2过表达进展期胃癌患者3年中位OS时间为23个月,3年OS率为33.4%;HER-2非过表达的进展期胃癌患者3年中位OS时间为29个月,3年OS率为44.6%,差异有统计学意义(P<0.05)。结论:AI辅助显微镜是一种用于胃癌原发灶不同区域HER-2评估实用可靠的工具,可以更为全面准确地评估胃癌的HER-2表达�Background and Aims:The expression of human epidermal growth factor receptor 2(HER-2)in gastric cancer exhibits significant heterogeneity,and comprehensive evaluation of HER-2 status in patients is advantageous for identifying potential beneficiaries of anti-HER-2 therapy.Artificial intelligence(AI)-assisted microscopy can integrate judgments on HER-2 status through extensive slide readings and reduce visual errors in human assessments.This study was performed to evaluate the practicality and feasibility of AI in evaluating HER-2 status in multiple regions of primary gastric cancer lesions.Methods:A total of 264 postoperative specimens from patients with advanced gastric cancer were evaluated for HER-2 expression in the same region using two methods:visual assessment with a conventional light microscope and AI-assisted microscopy.The accuracy of AI in HER-2 evaluation in gastric cancer was evaluated.Additionally,using AI-assisted microscopy,HER-2 expression in other regions of the primary lesion in the gastric cancer patients was assessed.The higher HER-2 score between the two regions was used as the final interpretation result.Furthermore,the associations of HER2 overexpression with clinicopathological features and postoperative survival of advanced gastric cancer patients were analyzed.Results:There was no statistically significant overall difference in HER-2 scores among visual assessment,AI-assisted microscopy,and the gold standard(P>0.05),but AI-assisted microscopy showed higher consistency compared to visual assessment with the gold standard(κ=0.86 vs.κ=0.81,P<0.05).The results of HER-2 evaluation in two different regions of the primary lesion in gastric cancer using AI showed inconsistency in 55 cases,with a discordance rate of 20.8%.The HER-2 overexpression rate was 29.9%in region 1 and 31.0%in region 2.The comprehensive evaluation of HER-2 overexpression rate,taking the higher score between the two regions,was 35.2%.The HER-2 overexpression rate based on comprehensive evaluation of HER-2 expression in

关 键 词：胃肿瘤 基因 erbB-1 人工智能 预后 

分 类 号：R735.2[医药卫生—肿瘤]