摘要
目的:基于术前CT构建预测肝门部胆管癌(pCCA)神经侵犯(PNI)的影像组学模型,并评价其效能。方法:回顾性分析本院2013年2月-2021年2月149例经病确诊的pCCA患者的临床资料,其中PNI组患者108例,无PNI组患者41例。采用R语言将所有患者按3:1比例随机分为训练集和验证集。在静脉期图像上,沿肿瘤边缘在所有层面上手动勾画3D感兴趣区(ROI),使用3D Slicer提取影像组学特征。采用组内相关系数(ICC)、相关性分析去除冗余特征,采用随机森林算法(RF)对所有临床、影像组学特征进行重要性排序,并选取前18个重要特征构建RF模型。使用准确性、敏感性、特异性及受试者操作特征(ROC)曲线评价模型效能。结果:在训练集中,RF模型的准确性、敏感性、特异性均为100%,ROC曲线下面积AUC为1;在验证集中,RF模型的准确性为70.3%,敏感性为59.3%,特异性为100%,AUC为0.846(0.713~0.979)。结论:基于增强CT图像建立的影像组学模型可用于术前无创性预测pCCA患者的PNI状态。
Objective:To construct a CT-based radiomics model for noninvasive predicting perineural invasion(PNI)of perihilar cholangiocarcinoma(pCCA)preoperatively,and to evaluate its discriminative abilities.Methods:Totally 149 patients with pCCA from February 2012 to October 2021 in the First Affiliated Hospital of Zhengzhou University were retrospectively enrolled in this study,including 108 patients in PNI positive group and 41 patients in PNI negative group.Based on the ratio of 3:1,all patients were randomly divided into training corhort and velidation cohort using the software R.The 3D region of interest(ROI)was segmented along the edge of the tumor on venous phase CT images and radiomics features were extracted using 3D slicer.The intraclass correlation efficient(ICC)and correlation analysis were applied to remove redundant features.To rank the importance of clinical and radiomics features,random forest algorithm(RF)was applied,and the first 18 ones were selected to build the RF model.To evaluate the performance of the model,the accuracy,sensitivity,specificity and receiver operating characteristic(ROC)curves were analyzed.Results:In the training cohort,the accuracy,sensitivity and specificity of the RF model were all 100%,and the AUC was 1.In the validation cohort,the accuracy of the RF model was 70.3%,the sensitivity was 59.3%,the specificity was 100%,and the AUC was 0.846(0.713~0.979).Conclusion:The enhanced CT-based radiomics model can be used for noninvasive prediction of PNI status preoperatively in pCCA patients.
作者
詹鹏超
刘珂衍
邱庆雅
沈佳宁
刘娜娜
王会霞
吕培杰
李臻
高剑波
ZHAN Peng-chao;LIU Ke-yan;QIU Qing-ya(Department of Radiology,the First Affiliated Hospital of Zhengzhou University,Zhengzhou 450052,China)
出处
《放射学实践》
CSCD
北大核心
2023年第7期910-915,共6页
Radiologic Practice
基金
河南省高等学校重点科研项目(22A320057)。
