含瘤周移行带影像组学模型预测肺腺癌病理分级  

Prediction of pathological grading of lung adenocarcinoma by comparing peritumoral with intratumoral radiomics models

作　　者：陈欢[1,2] 梁明柱[1] 雷益[2] 李昕[3] 谢传淼 张兰军[5] 柳学国[1] CHEN Huan;LIANG Ming-zhu;LEI Yi(Department of Radiology,the Fifth Hospital of Zhongshan University,Guangdong 519000,China)

机构地区：[1]中山大学附属第五医院放射科,广东519000 [2]深圳大学附属第一医院放射科,广东518000 [3]GE医疗生命医学部,广州510000 [4]中山大学肿瘤防治中心放射科,广州510000 [5]中山大学肿瘤防治中心胸外科,广州510000

出　　处：《放射学实践》2020年第4期478-483,共6页Radiologic Practice

摘　　要：目的:探讨含肿瘤外缘5mm移行带的影像组学模型对预测肺腺癌病理分级的诊断效能。方法:回顾性搜集173例经手术病理证实的肺腺癌患者的胸部增强CT薄层图像及临床病理资料。其中女96例,男77例;年龄33~84岁,平均(60.0±1.2)岁;病灶直径6.0~30.0 mm,平均(18.0±1.6)mm;实性结节102例,亚实性结节71例;Ⅰ期134例,Ⅱ期18例,Ⅲ期21例;病理分级为1级51例,2级114例,3级8例。病理分级按最主要亚型分为1级(原位腺癌、微浸润腺癌和贴壁为主型腺癌)、2级(包括腺泡或乳头为主型腺癌)和3级(包括实性或微乳头为主型腺癌)。根据结节可见边缘(瘤内组)和自动外扩5 mm(含瘤周组)对结节进行分割并使用软件自动提取影像组学特征。训练集与验证集比例为7∶3,使用随机森林构建影像组学模型,并利用混淆矩阵的准确性来评价瘤内及含瘤周影像组学模型预测肺腺癌病理分级的效能。结果:每例患者共提取病灶的385个影像组学特征,其中瘤内组中8个影像组学特征和含瘤周组中12个影像组学特征与肺腺癌的病理分级显著相关。瘤内影像组学模型及含瘤周影像组学模型预测肺腺癌病理分级的准确性在训练集为90.83%vs.92.61%(P>0.05),在验证组为90.74%vs.94.44%(P>0.05)。结论:与瘤内组影像组学模型相比,含5mm瘤周影像组学模型可略提高对肺腺癌病理分级预测的准确性。Objective:The purpose of this study was to explore the diagnostic efficacy in prediction of pathological grading of lung adenocarcinoma by comparing the peritumoral radiomics models of 5 mm transitional zone with intratumoral ones.Methods:The CT images and clinical pathological data of 173 patients with lung adenocarcinoma proved by operation and pathology were retrospectively collected and analyzed.There were 96 females and 77 males,with age of 33~84(60.0±1.2)years.The size of tumor ranged from 6 to 30mm with mean of(18.0±1.6)mm.There were 102 solid,71 non-solid nodules;clinical stageⅠin 134 cases,stageⅡin 18 cases,and stageⅢin 21 cases.Pathological grade was classified as grade 1,2 and 3 according to the most predominant subtype proportion.Grade 1,2 and 3 were found in 51,114 and 8 cases,respectivly.Radiomics features of pulmonary nodules were automatically extracted by software according to the visible edge(intratumoral group)and the 5mm auto-extending transitional zone(peritumoral group).The ratio of training set to testing set was 7∶3.The radiomics models were formed by random forest model.Then the accuracy of confusion matrix was used to evaluate the effectiveness of intratumoral and peritumoral radiomics models in predicting the pathological grading of lung adenocarcinoma.Results:A total of 385 radiomics features were extracted from the lesion in each subject.Among which there were 8 features of intratumoral group and 12 features of peritumoral group significantly associated with pathological grading.The accuracy of intratumoral and peritumoral radiomics model of predicting pathological grading was 92.61%vs.90.83%(P>0.05)in training set,and 94.44%vs.90.74%(P>0.05)in testing set.Conclusion:5mm peritumoral radiomics model can improve the accuracy of predicting pathological grading of lung adenocarcinoma compared with intratumoral radiomics.

关 键 词：体层摄影术 X线计算机 影像组学 肺肿瘤 腺癌 病理分级 移行带 

分 类 号：R445.2[医药卫生—影像医学与核医学] R734.2[医药卫生—诊断学]