CT放射组学联合血清肿瘤标志物对不确定性质肺结节恶性风险的预测价值  

Prognostic value of CT radiomics combined with serum tumor markers for malignant risk of uncertain pulmonary nodules

作　　者：曹冠杰[1] 史志涛[1] 王彩华[2] 孙占国[1] 陈月芹[1] 靳超[3] 李新勃 CAO Guanjie;SHI Zhitao;WANG Caihua;SUN Zhanguo;CHEN Yueqin;JIN Chao;LI Xinbo(Department of Imaging,Affiliated Hospital of Jining Medical College,Jining,Shandong 272007,China;Department of Otolaryngology,Affiliated Hospital of Jining Medical College,Jining,Shandong 272007,China;Department of Laboratory Medicine,Affiliated Hospital of Jining Medical College,Jining,Shandong 272007,China)

机构地区：[1]济宁医学院附属医院影像科,山东济宁272007 [2]济宁医学院附属医院耳鼻喉科,山东济宁272007 [3]济宁医学院附属医院检验科,山东济宁272007

出　　处：《临床肺科杂志》2023年第9期1321-1326,共6页Journal of Clinical Pulmonary Medicine

基　　金：山东省医药卫生科技发展计划项目(No.202009011151);济宁市重点研发项目(No.2021YXNS128)。

摘　　要：目的分析CT放射组学联合血清肿瘤标志物对不确定性质肺结节(IPN)恶性风险的预测价值。方法选取2019年3月至2023年3月诊治的IPN患者238例,检测血清肿瘤标志物水平。根据CT放射组学,计算基于结节形状、大小和质地的放射组学评分。使用Logistic回归(LR)和随机森林(RF)开发预测模型,与目前的风险评估标准(Mayo)进行比较。使用偏倚校正临床净重新分类指数(cNRI)确定IPN风险重新分类。结果LR和RF建模均显示年龄、放射组学、CYFRA 21-1和CEA是IPN恶性风险的最强预测因子。与Mayo相比,LR和RF模型具有更高的诊断准确性,LR模型的AUC为0.764,RF模型的AUC为0.731。与Mayo模型相比,LR和RF模型恶性结节重新分类的cNRI分别为为0.21(0.20,0.23)和0.21(0.19,0.23)。结论放射组学、CYFRA 21-1、CEA和患者年龄联合模型具有比Mayo更高的IPN诊断准确性。Objective To analyze the predictive value of CT imaging features combined with serum tumor markers for malignant risk of uncertain pulmonary nodules(IPNs).Methods A total of 238 IPN patients diagnosed and treated between March 2019 and March 2023 were selected to detect the levels of serum tumor markers.Radiomic scores based on the shape,size,and texture of nodules were calculated based on CT radiomics.logistic regression(LR)and Random Forest(RF),and were used to develop prediction models compared with current risk assessment criteria(Mayo).The bias-adjusted Clinical Net reclassification Index(cNRI)was used to determine IPN risk reclassification.Results Both LR and RF modeling showed that age,radiomics,CYFRA 21-1,and CEA were the strongest predictors of IPN malignancy risk.Compared with Mayo,the LR and RF models had higher diagnostic accuracy,with an AUC of 0.764 for the LR model and 0.731 for the RF model.Compared with Mayo model,the cNRI of LR and RF models for malignant nodule reclassification were 0.21(0.20,0.23)and 0.21(0.19,0.23).Conclusion The combined model of radiomics,CYFRA 21-1,CEA,and patient age have higher diagnostic accuracy for IPN than Mayo.

关 键 词：不确定性质肺结节 肺癌 肿瘤标志物 放射组学 预测模型 

分 类 号：R563[医药卫生—呼吸系统]