基于薄层增强CT图像探究影像组学在肺结节良恶性诊断价值  

Exploring the Value Of Radiomics in the Diagnosis of Benign and Malignant Pulmonary Nodules Based on Thin-Section Enhanced CT Images

作　　者：沈天赐 宋鑫洋 胡翔宇 郭建峰 杨峰 SHEN Tian-ci;SONG Xin-yang;HU Xiang-yu;GUO Jian-feng;YANG Feng(Department of Radiology,Xiangyang No.1 People's Hospital Affiliated to Hubei University of Medicine,Xiangyang 441000,Hubei Province,China;Department of Respiration,Xiangyang No.1 People's Hospital Affiliated to Hubei University of Medicine,Xiangyang 441000,Hubei Province,China)

机构地区：[1]湖北医药学院附属襄阳市第一人民医院放射科,湖北襄阳441000 [2]湖北医药学院附属襄阳市第一人民医院呼吸科,湖北襄阳441000

出　　处：《中国CT和MRI杂志》2025年第4期75-78,共4页Chinese Journal of CT and MRI

基　　金：湖北省“323”攻坚行动襄阳市第一人民医院重点专项科研基金(XYY2022-323);襄阳市第一人民医院医院科技创新项目(XYY2023SD18);2022年湖北医药学院研究生科技创新项目(YC2022049);2023年湖北医药学院研究生科技创新项目(YC2023050)。

摘　　要：目的基于临床及薄层增强CT影像组学构建联合模型,评估其预测诊断肺结节患者良恶性的价值。方法回顾性分析116例(良性52例,恶性64例)经病理证实的肺结节患者,将患者采用分层抽样的方式按照7:3的比例分为训练组和验证组,沿结节边缘逐层提取每位患者薄层增强(动脉期及静脉期)图像肺结节区域的感兴趣区(regions of interest,ROI),采用3Dslicer软件提取影像组学纹理特征,使用LASSO回归算法对影像组学特征进行特征筛选及降维,选择非零变量构建影像组学特征模型。结合独立的临床危险因素采用多元Logistics回归建立影像组学列线图,列线图的准确率和诊断效能在训练集中进行评估,随后在验证集中进行验证,最后通过决策曲线分析评估列线图在临床实践中的应用价值。结果基于临床特征模型在训练集(AUC=0.81,95%CI 0.73-0.90)及验证集(AUC=0.85,95%CI 0.73-0.98)对肺结节良恶性诊断均有所欠佳,筛选出9个影像组学纹理特征与肺结节良恶性相关依据回归系数建立Rad-Score特征模型在训练集及验证集AUC分别为0.91(95%CI 0.84-0.98),0.90(95%CI 0.80-1.00),联合影像组学及临床特征列线图模型在训练集(0.94,95%CI 0.89-0.99)和验证集(0.98,95%CI 0.94-1.00)均表现优异,DCA分析结果表明影像组学的加入可以使患者获益。结论联合临床特征及增强CT影像组学建立的列线图模型具有良好预测肺结节良恶性的效能。Objective To construct a joint model based on clinical and thin-section enhanced CT radiomics to assess its value in predicting the diagnosis of benign and malignant pulmonary nodules in patients.Methods We retrospectively analyzed 116 patients(52 benign and 64 malignant)with pathologically confirmed pulmonary nodules,divided the patients into training and validation groups by stratified sampling in the ratio of 7:3,extracted regions of interest(ROI)of pulmonary nodule regions from thin-section enhanced(arterial and venous)images along the edge of the nodule of each patient stratified by stratification,and used The 3D slicer software was used to extract the texture features,and the LASSO regression algorithm was used to filter and reduce the dimensionality of the radiomics features,and non-zero variables were selected to construct the radiomics feature model.The diagnostic performance of the column line maps were evaluated in the training cohort and validated in the validation cohort,and finally the clinical utility of the column line maps was evaluated by decision curve analysis.Results Based on the clinical feature model in the training set(AUC=0.81,95%CI 0.73-0.90)and the validation set(AUC=0.85,95%CI 0.73-0.98),the diagnosis of benign and malignant pulmonary nodules was slightly poor,and nine radiomics texture features were selected to correlate with benign and malignant pulmonary nodules based on the regression coefficients to establish the Rad-Score feature model in the training and validation sets.The AUC of the validation set was 0.91(95%CI 0.84-0.98)and 0.90(95%CI 0.80-1.00),respectively,and the combined radiomics and clinical features line graph model performed well in both the training(0.94,95%CI 0.89-0.99)and validation sets(0.98,95%CI 0.94-1.00),and the DCA analysis The results suggest that the inclusion of radiomics can benefit patients.Conclusion The column line graph model established by combining clinical features and enhanced CT radiomics has good efficacy in predicting the benignity and maligna

关 键 词：肺结节 体层摄影术 X线计算机 影像组学 

分 类 号：R563[医药卫生—呼吸系统]