基于非小细胞肺癌双能CT表现及影像组学列线图模型预测其血管生成拟态  

Nomogram based on dual-energy CT findings and radiomics for predicting vasculogenic mimicry in non-small cell lung cancer

作　　者：李阳[1,2] 杨昭 李淑华 赵楠楠[1,2] 张舒妮 杨静茹 周辉 李伟 蒋明宽 谢宗玉 LI Yang;YANG Zhao;LI Shuhua;ZHAO Nannan;ZHANG Shuni;YANG Jingru;ZHOU Hui;LI Wei;JIANG Mingkuan;XIE Zongyu(Department of Radiology,the First Affiliated Hospital of Bengbu Medical College,Bengbu 233004,China;Department of Postgraduate,Bengbu Medical College,Bengbu 233000,China;Department of Medical Imaging,Fengyang County People's Hospital,Chuzhou 233100,China;Respiratory DiseasesClinical Medical Research Center of Anhui Province,Bengbu 233004,China)

机构地区：[1]蚌埠医学院第一附属医院放射科,安徽蚌埠233004 [2]蚌埠医学院研究生院,安徽蚌埠233000 [3]凤阳县人民医院医学影像科,安徽滁州233100 [4]安徽省呼吸系统疾病(肿瘤)临床医学研究中心,安徽蚌埠233004

出　　处：《中国医学影像技术》2023年第5期684-689,共6页Chinese Journal of Medical Imaging Technology

基　　金：安徽省重点研究与开发计划项目(2022e07020033);蚌埠医学院自然科学重点项目(2021byzd091);滁州市科技计划项目(2022ZD007)。

摘　　要：目的基于非小细胞肺癌(NSCLC)双能CT(DECT)表现及影像组学构建联合列线图模型,分析其预测NSCLC血管生成拟态(VM)的价值。方法回顾性分析137例经手术病理证实的单发NSCLC患者,以7∶3比例将其分为训练集[n=95,37例VM(+)、58例VM(-)]和验证集[n=42,19例VM(+)、23例VM(-)]。基于肺窗CT提取及筛选最优影像组学特征,计算影像组学评分。以单因素分析及多因素logistic回归分析筛选NSCLC表达VM的独立预测因素,分别以之构建临床、能谱及影像组学模型;基于独立预测因素构建联合列线图模型。采用受试者工作特征曲线评估各模型预测NSCLC VM的效能,以校准曲线分析模型的拟合度,以决策曲线分析评估模型的临床获益。结果最终筛选出6个最优影像组学特征。病灶最大径、毛刺征、CT_(140 keV)及影像组学评分为NSCLC VM的独立预测因素(OR=2.25、9.69、0.99、-14.44,P均<0.05)。临床、能谱及影像组学模型预测验证集NSCLC VM的曲线下面积(AUC)分别为0.83、0.85、0.87,均低于联合列线图模型(AUC=0.95,Z=2.14、2.10、2.07,P均<0.05)。联合列线图模型预测结果与实际结果的一致性较好,且其临床获益较高。结论基于DECT及影像组学构建的联合列线图模型能可有效预测NSCLC VM。Objective To construct a combined nomogram model based on dual energy CT(DECT)findings and radiomics,and to analyze its value for predicting vasculogenic mimicry(VM)in non-small cell lung cancer(NSCLC).Methods Totally 137 patients with surgical pathologically confirmed single NSCLC were enrolled and divided into training set(n=95,37 VM[+]and 58 VM[-])and validation set(n=42,19 VM[+]and 23 VM[-])at the ratio of 7∶3.Based on lung window CT,the optimal radiomics features were extracted and screened.Univariate analysis and multivariate logistic regression analysis were used to screen independent predictors of VM in NSCLC.Clinical,DECT and radiomics models were conducted,respectively,as well as a combined nomogram model based on independent predictors.Receiver operating characteristic curve was used to evaluate the efficacy of the above models for predicting VM in NSCLC.The fits of the models were explored using calibration curves,and the clinical benefits of the models were assessed using decision curve analysis.Results Six optimal radiomics features were finally screened for calculation of the radiomics score.The maximum diameter,burr sign,CT_(140 keV) and radiomics score of lesion were independent predictors of VM in NSCLC(OR=2.25,9.69,0.99,-14.44,all P<0.05).The area under the curve(AUC)of the clinical,DECT and radiomics model for predicting VM in NSCLC in validation set was 0.83,0.85 and 0.87,respectively,lower than that of combined nomogram model(AUC=0.94,Z=2.14,2.10,2.07,all P<0.05).The predicted results of combined nomogram model were in good agreement with actual results,while the combined nomogram model had the higher clinical benefit.Conclusion Combined nomogram model based on DECT findings and radiomics could effectively predict VM in NSCLC.

关 键 词：肺肿瘤 体层摄影术 X线计算机 影像组学 血管生成拟态 

分 类 号：R734.2[医药卫生—肿瘤] R814.42[医药卫生—临床医学]