食管鳞癌淋巴结转移的影像组学与预后的关联

Association between Radiomics Related to Lymph Node Metastasis and Prognosis of Esophageal Squamous Cell Carcinoma

下载PDF

导出

摘要目的探讨使用卷积神经网络(CNN)提取的食管鳞癌(ESCC)淋巴结转移相关的影像组学特征与ESCC预后的关联。方法回顾性分析308例ESCC患者的临床资料,其中154例有淋巴结转移、154例无淋巴结转移,按2∶1比例将其分为训练集和验证集。通过MRIcroGL软件标记CT影像中的淋巴结,使用CNN分割提取ESCC淋巴结影像特征,通过LASSO回归和随机森林筛选与ESCC淋巴结转移相关的影像组学特征并构建预测模型,采用Cox回归进行特征选择,建立影像组学标签,分析其与ESCC预后的关联,进而构建列线图评价模型预测能力。结果使用CNN自动提取到999个影像组学特征值;使用LASSO回归筛选出的19个特征,在全模型、训练集和验证集中构建模型的曲线下面积(AUC)(95%CI)分别为0.747(0.694,0.801)、0.751(0.686,0.817)和0.766(0.672,0.860);使用随机森林筛选出9个特征的预测模型中AUC (95%CI)分别为0.692(0.633,0.751)、0.683(0.610,0.755)和0.723(0.624,0.822)。多因素Cox回归分析显示,高风险的影像组学标签与ESCC不良的预后相关(P<0.05)。列线图结果显示,影像组学标签能很好地预测ESCC预后,总生存率C指数(95%CI)为0.710(0.670,0.749),无病进展生存率C指数(95%CI)为0.775(0.746,0.804)。结论基于CNN和CT影像组学构建的影像组学标签对预测ESCC预后具有很高的价值。 Objective To explore the relationship between the radiomics features associated with lymph node metastasis(LNM)of esophageal squamous cell carcinoma(ESCC)and the prognosis of ESCC by using convolutional neural networks(CNN)screening.Methods This study enrolled 308 patients with pathologically confirmed advanced ESCC,including 154 with LNM and 154 without LNM.All enrolled patients were randomly divided into a training cohort and a testing cohort at a ratio of 2∶1.The lymph nodes in CT images are marked by MRIcroGL software,then radiomics features of ESCC lymph nodes were segmented and extracted by CNN.Radiomics features associated with ESCC LNM were screened by LASSO method and the random forest,these features were used to construct a prediction model.Using Cox regression to feature selection,meanwhile establishing radiomics signature based on those selected features.And analyzing the association between radiomics signature and ESCC prognosis,and then constructing the nomogram to evaluate the predictive ability of the model.Results By using CNN,999 radiomics signature feature values are automatically extracted.19 features were screened out by LASSO,using these features to build predictive models of ESCC LNM,with 0.747(95%CI:0.694,0.801)in the whole cohort,0.751(95%CI:0.686,0.817)in the training cohort and 0.766(95%CI:0.672,0.860)in the testing cohort.9 features were screened out by the random forest and the prediction model of ESCC LNM was constructed,which were 0.692(95%CI:0.633,0.751),0.683(95%CI:0.610,0.755)and 0.723(95%CI:0.624,0.822)in the whole cohort,training cohort and testing cohort,respectively.Multivariate Cox regression analysis showed that high-risk radiomics signature was associated with poor prognosis of ESCC(P<0.05).The Nomogram results showed that radiomics signature could more precisely predict the prognosis of ESCC.The C-index of the nomogram for overall survival was 0.710(95%CI:0.670,0.749)and disease free survival was 0.775(95%CI:0.746,0.804).Conclusion The radiomics signature constructed ba

作者陈小琪徐华霞王小玥李锐进邱明链邱模良吴凯明胡志坚 CHEN Xiaoqi;XU Huaxia;WANG Xiaoyue;LI Ruijin;QIU Minglian;QIU Moliang;WU Kaiming;HU Zhijian(Public Health School of Fujian Medical University,Fuzhou 350122,China;School of Basic Medical Sciences Fujian Medical University,Fuzhou 350122,China;Department of Thoracic Surgery,The First Affiliated Hospital of Fujian Medical University,Fuzhou 350004,China;Department of Imaging,Fuzhou First Hospital Affiliated to Fujian Medical University,Fuzhou 350009,China;Department of Imaging,The First Aftiliated Hospital of Fujian Medical University,Fuzhou 350004,China;Epidemiology and Health Statistics,Public Health School of Fujian Medical University,Fuzhou 350122,China)