基于不同训练方法的2种人工智能系统在染色放大内镜下识别早期胃癌的能力比较  

Comparison of the ability of two artificial intelligence systems based on different training methods to diagnose early gastric cancer under magnifying image-enhanced endoscopy

作　　者：朱益洁 吴练练 何鑫琦 李艳霞[1] 周巍[1] 张军[1] 蒋逍达[1] 于红刚[1] Zhu Yijie;Wu Lianlian;He Xinqi;Li Yanxia;Zhou Wei;Zhang Jun;Jiang Xiaoda;Yu Honggang(Department of Gastroenterology,Renmin Hospital of Wuhan University,Hubei Key Laboratory of Digestive Diseases,Hubei Clinical Research Center for Minimally Invasive Diagnosis and Treatment of Digestive Diseases,Wuhan 430060,China)

机构地区：[1]武汉大学人民医院消化内科,消化系统疾病湖北省重点实验室,湖北省消化疾病微创诊治医学临床研究中心,武汉430060

出　　处：《中华消化杂志》2022年第7期433-438,共6页Chinese Journal of Digestion

基　　金：湖北省技术创新重大项目(CXZD2019000034);湖北省卫生健康委员会创新团队项目(WJ2021C003)。

摘　　要：目的比较随机裁剪图片深度卷积神经网络识别模型(DCNN-C)和整张图片深度卷积神经网络识别模型(DCNN-W)2种基于不同训练方法的人工智能系统在染色放大内镜下辅助识别早期胃癌的能力。方法回顾性收集武汉大学人民医院内镜中心窄带成像和蓝激光成像2种染色放大内镜模式下的早期胃癌或非癌图片和视频片段,用于DCNN-C和DCNN-W的训练集和测试集。比较DCNN-C和DCNN-W在图片测试集中,以及DCNN-C、DCNN-W和3名高年资内镜医师(平均水平)在视频测试集中识别早期胃癌的能力。统计学方法采用配对卡方检验和卡方检验。观察者间的一致性以Cohen′s Kappa统计系数(Kappa值)表示。结果在图片测试集中,DCNN-C诊断早期胃癌的准确度、灵敏度、特异度、阳性预测值分别为94.97%(1133/1193)、97.12%(202/208)、94.52%(931/985)、78.91%(202/256),分别高于DCNN-W的86.84%(1036/1193)、92.79%(193/208)、85.58%(843/985)、57.61%(193/335),差异均有统计学意义(χ^(2)=4.82、4.63、61.04、29.69,P=0.028、=0.035、<0.001、<0.001)。在视频测试集中,高年资内镜医师诊断早期胃癌的准确度、特异度和阳性预测值分别为67.67%、60.42%、53.37%,分别低于DCNN-C的93.00%、92.19%、87.18%,差异均有统计学意义(χ^(2)=20.83、16.41、11.61,P<0.001、<0.001、=0.001);DCNN-C诊断早期胃癌的准确度、特异度和阳性预测值分别高于DCNN-W的79.00%、70.31%、64.15%,差异均有统计学意义(χ^(2)=7.04、8.45、6.18,P=0.007、0.003、0.013);高年资内镜医师诊断早期胃癌的准确度、特异度、阳性预测值与DCNN-W比较差异均无统计学意义(均P>0.05);高年资内镜医师、DCNN-W和DCNN-C诊断早期胃癌的灵敏度分别为80.56%、94.44%、94.44%,差异均无统计学意义(均P>0.05)。一致性分析结果显示,高年资内镜医师与金标准的一致性一般至中等(Kappa值=0.259、0.532、0.329),DCNN-W与金标准的一致性中等(Kappa值=0.587),DCNN-C与金标Objective To compare the ability of deep convolutional neural network-crop(DCNN-C)and deep convolutional neural network-whole(DCNN-W),2 artificial intelligence systems based on different training methods to dignose early gastric cancer(EGC)diagnosis under magnifying image-enhanced endoscopy(M-IEE).Methods The images and video clips of EGC and non-cancerous lesions under M-IEE under narrow band imaging or blue laser imaging mode were retrospectively collected in the Endoscopy Center of Renmin Hospital of Wuhan University,for the training set and test set for DCNN-C and DCNN-W.The ability of DCNN-C and DCNN-W in EGC identity in image test set were compared.The ability of DCNN-C,DCNN-W and 3 senior endoscopists(average performance)in EGC identity in video test set were also compared.Paired Chi-squared test and Chi-squared test were used for statistical analysis.Inter-observer agreement was expressed as Cohen′s Kappa statistical coefficient(Kappa value).Results In the image test set,the accuracy,sensitivity,specificity and positive predictive value of DCNN-C in EGC diagnosis were 94.97%(1133/1193),97.12%(202/208),94.52%(931/985),and 78.91%(202/256),respectively,which were higher than those of DCNN-W(86.84%,1036/1193;92.79%,193/208;85.58%,843/985 and 57.61%,193/335),and the differences were statistically significant(χ^(2)=4.82,4.63,61.04 and 29.69,P=0.028,=0.035,<0.001 and<0.001).In the video test set,the accuracy,specificity and positive predictive value of senior endoscopists in EGC diagnosis were 67.67%,60.42%,and 53.37%,respectively,which were lower than those of DCNN-C(93.00%,92.19%and 87.18%),and the differences were statistically significant(χ^(2)=20.83,16.41 and 11.61,P<0.001,<0.001 and=0.001).The accuracy,specificity and positive predictive value of DCNN-C in EGC diagnosis were higher than those of DCNN-W(79.00%,70.31%and 64.15%,respectively),and the differences were statistically significant(χ^(2)=7.04,8.45 and 6.18,P=0.007,0.003 and 0.013).There were no significant differences in accuracy,specificity a

关 键 词：早期胃癌 人工智能 染色放大内镜 卷积神经网络 

分 类 号：R735.2[医药卫生—肿瘤]