基于CT平扫图像的纹理分析区分胃癌HER2表达的可行性研究  

Discrimination of HER2 expression in gastric cancer by texture analysis based on plain CT images:a feasibility study

作　　者：张笑[1] 黄伟[1] 宋彬[2] ZHANG Xiao;HUANG Wei;SONG Bin(Department of Radiology,People's Hospital of Leshan City,Leshan,Sichuan 614000,P.R.China;Department of Radiology,West China Hospital,Sichuan University,Chengdu 610041,P.R.China)

机构地区：[1]四川省乐山市人民医院放射影像科,四川乐山614000 [2]四川大学华西医院放射科,成都610041

出　　处：《中国普外基础与临床杂志》2021年第9期1221-1226,共6页Chinese Journal of Bases and Clinics In General Surgery

摘　　要：目的探讨对胃癌CT平扫图像进行纹理分析与胃癌HER2表达的相关性。方法回顾性收集2017年1月至2021年1月期间在乐山市人民医院行上腹部和(或)全腹部CT平扫检查、行手术治疗且术后检测了胃癌组织HER2表达状况的62例胃癌患者的临床资料,其中男45例,女17例;肿瘤的Lauren分型:肠型18例,弥漫型30例,混合型14例;HER2表达阴性52例[年龄(63.54±10.32)岁],HER2表达阳性10例[年龄(61.70±11.70)岁]。使用MaZda 4.6版本软件对病例的CT平扫图像在MaZda模块中做图像均一化、兴趣区勾画、纹理特征提取和纹理特征选择,在B11模块中进行纹理特征判别和误判率分析。结果HER2表达状态与患者年龄、性别和肿瘤的Lauren分型无相关性(P>0.05)。非线性判别分析/人工神经网络(NDA/ANN)、线性判别分析/1-最近邻(LDA/1-NN)、主成分分析/1-最近邻(PCA/1-NN)和原始数据分析/1-最近邻(RDA/1-NN)各分析方法能较好地将胃癌的CT平扫纹理特征参数和HER2表达水平对应在一起。结论基于CT平扫图像的纹理分析具有无创检测胃癌HER2表达状态的潜力;综合效能最好的纹理判别方法是NDA/ANN和LDA/1-NN。Objective To explore the correlation between the texture features of gastric cancer plain CT images and the expression of HER2.Methods A retrospective collection the datas of 62 patients with gastric cancer who underwent CT scans of the upper abdomen and(or)the whole abdomen from January 2017 to January 2021 in Leshan City People’s Hospital.The treatment method was surgery.The HER2 expression of gastric cancer tissue was detected after the operation.There were 45 male patients and 17 female patients.Lauren classification:18 cases of intestinal type,30 cases of diffuse type,and 14 cases of mixed type.Fifty-two cases were HER2 expression negative[age:(63.54±10.32)years],and 10 cases were HER2 expression positive[age:(61.70±11.70)years].The MaZda module in the MaZda 4.6 version software was used to perform the image normalization,interest area delineation,texture feature extraction,and texture feature selection on the CT plain scan image,and perform texture feature discrimination and misjudgment rate analysis in the B11 module.Results There was no correlation between HER2 expression and age,gender of patients and Lauren classification of tumors(P>0.05).The analysis methods of nonlinear discriminant analysis(NDA)/artificial neural network(ANN),linear discriminant analysis(LDA)/1-nearest-neighbor(1-NN),principal component analysis(PCA)/1-NN,and raw-data analysis(RDA)/1-NN can better correspond to the CT plain scan texture feature parameters of gastric cancer and the expression level of HER2.Conclusions Texture analysis based on CT plain images has the potential to non-invasively detect the HER2 expression in gastric cancer.The best comprehensive performance texture discrimination method is NDA/ANN and LDA/1-NN.

关 键 词：胃癌 人表皮生长因子受体2 纹理分析 MaZda软件 误判率 

分 类 号：R735.2[医药卫生—肿瘤]