摘要
目的评估多期相融合CT影像组学鉴别小细胞肺癌(SCLC)与非小细胞肺癌(NSCLC)的价值并建立鉴别SCLC与NSCLC的预测模型。方法回顾性搜集2018年12月至2022年12月经病理学确诊的肺癌患者肺部CT(肺窗、纵隔窗、动脉期、静脉期薄层图像)及临床资料。提取影像资料和临床资料中最有价值的特征,分别用以构建多期相融合CT影像组学模型(为了突出多期相融合的有效性,将多期相融合影像组学模型与单一期相影像组学模型作对比),临床模型及联合模型,模型的构建使用三种较为常用的机器学习算法:Logistic回归(LR)、k-近邻(KNN)及多层感知器(MLP)。结果结合LR机器学习算法构建的联合模型表现出最佳的效能[训练集曲线下面积(AUC)=0.946,测试集AUC=0.911]。融合期相影像组学模型的AUC(训练集:0.849,测试集:0.847)比单一期相影像组学模型更高且具有正向改善力[连续净重分类改善度(NRI),综合判别改善度(IDI)均>0],同时,校准曲线显示,融合期相影像组学模型预测结果与真实结果之间一致性较高。结论多期相融合CT影像组学在鉴别SCLC与NSCLC中具备较高的价值;融合期相影像组学模型诊断效能比单一期相影像组学模型更加出色;在联合神经元特异性烯醇化酶(NSE)和胃泌素释放肽前体(ProGRP)两个临床独立危险因素后,融合期相影像组学模型的效能有所改善,可为鉴别SCLC和NSCLC提供一种客观全面、准确快捷、无创安全及高成本效益的方法;LR机器学习算法比KNN及MLP机器学习算法表现更好。
Objective To evaluate the value of multistage fusion CT radiomics to identify small cell lung cancer(SCLC)and non-small cell cancer(NSCLC)and establish a predictive model to identify SCLC from NSCLC.Methods Pulmonary CT data(Thin-layer images of lung window,mediastinal window,arterial stage,and venous stage)and clinical data of pathologically confirmed lung cancer patients were collected from December 2018 to December 2022.Extract image data and clinical data of the most valuable features,respectively to build multiple fusion CT radiomics model(To highlight the effectiveness of multiphase fusion,the multiphase fusion radiomics model was compared with a single phaseradiomics model),clinical model and joint model,model construction using three commonly used machine learning algorithm:logistic regression(LR),k-nearest neighbors(KNN)and multilayer perceptron(MLP).Results The combined model built with LR machine learning algorithm showed the best performance(training set AUC=0.946,test set AUC=0.911).The AUC(training set:0.849,test set:0.847)of the fusion phaseradiomics model was higher than that of the single phase radiomics model and had positive improvement(NRI,IDI>0).Meanwhile,the calibration curve showed that the predicted results of the fusion phase radiomics model were more consistent with the real results.Conclusion Multi-phase fusion CT imaging has a high value in differentiating SCLC from NSCLC.The diagnostic efficiency of fusion phase radiomics model is better than that of single phase radiomics model.In combination with NSE and ProGRP clinical independent risk factors,the efficiency of the fusion phase radiomics model has been improved.It can provide an objective,comprehensive,accurate,rapid,non-invasive and cost-effective method for the differentiation of SCLC and NSCLC.In this study,LR machine learning algorithm performs better than KNN and MLP machine learning algorithm.
作者
杨福军
沈芳
唐艳隆
李祥宇
张蝶
王文丽
王鹏
苏宗友
YANG Fujun;SHEN Fang;TANG Yanlong(Western Yunnan Regional Medical Center of Dali People's Hospital,Dali,Yunnan Province 671014,P.R.China)
出处
《临床放射学杂志》
北大核心
2024年第10期1722-1729,共8页
Journal of Clinical Radiology
