王平,裴旭,殷小平,任嘉梁,郭建党,赵珍珍,赵莹佳.基于增强CT影像组学模型鉴别肾透明细胞癌与非透明细胞癌[J].中国医学影像技术,2019,35(11):1689~1692
基于增强CT影像组学模型鉴别肾透明细胞癌与非透明细胞癌
Radiomics models based on enhanced CT for identifying renal clear cell carcinoma and non-clear cell carcinoma
投稿时间:2019-01-15  修订日期:2019-07-03
DOI:10.13929/j.1003-3289.201901099
中文关键词:  肾肿瘤  影像组学  体层摄影术,X线计算机
英文关键词:kidney neoplasms  radiomics  tomography, X-ray computed
基金项目:河北省研究生创新资助项目(hbu2019ss036)、保定市科技计划项目(18ZF182)。
作者单位E-mail
王平 河北大学附属医院CT-MRI室, 河北 保定 071000
河北大学(医学院), 河北 保定 071000 
 
裴旭 河北大学附属医院CT-MRI室, 河北 保定 071000
河北大学(医学院), 河北 保定 071000 
 
殷小平 河北大学附属医院CT-MRI室, 河北 保定 071000
河北省肿瘤放化疗机制与研究重点实验室, 河北 保定 071000 
yinxiaoping78@sina.com 
任嘉梁 通用电气药业(上海)有限公司, 上海 210000  
郭建党 河北大学附属医院CT-MRI室, 河北 保定 071000  
赵珍珍 河北大学附属医院CT-MRI室, 河北 保定 071000  
赵莹佳 河北大学附属医院CT-MRI室, 河北 保定 071000  
摘要点击次数: 1727
全文下载次数: 600
中文摘要:
      目的 建立基于增强CT的影像组学模型,评估其鉴别肾透明细胞癌(ccRCC)与非透明细胞癌(non-ccRCC)的应用价值。方法 将147例ccRCC及32例non-ccRCC患者随机分为训练集125例和测试集54例。将所有患者的增强CT资料导入ITK-SNAP软件,手动勾画ROI,获得16个特征,分别建立基于特征的随机森林(RF)模型和逻辑回归(LR)模型,采用ROC曲线观察模型对ccRCC的诊断效能。结果 训练集RF模型诊断ccRCC的AUC为0.96(P<0.05),特异度为1.00,敏感度0.83;LR模型诊断ccRCC的AUC为0.96(P<0.05),特异度为1.00,敏感度为0.83。测试集RF模型诊断ccRCC的AUC为0.96(P<0.05),特异度为1.00,敏感度为0.89;LR模型诊断ccRCC的AUC为0.88(P<0.05),特异度为0.90,敏感度为0.77。结论 基于增强CT影像组学模型可用于鉴别ccRCC与non-ccRCC;RF模型诊断价值较LR模型更高。
英文摘要:
      Objective To establish radiomics models based on enhanced CT, and to explore the value of the models for distinguishing renal clear cell carcinoma (ccRCC) and non-clear cell renal cell carcinoma (non-ccRCC). Methods Totally 147 patients with ccRCC and 32 patients with non-ccRCC were randomly divided into training set (n=125) and testing set (n=54). Enhanced CT data were imported into ITK-SNAP software, and ROI was manually delineated to obtain 16 features. Random Forest (RF) model and Logistic Regression (LR) model based on features were established, respectively. ROC curve was used to observe the diagnostic efficiency of the models for ccRCC. Results In the training set, RF model diagnosed ccRCC with AUC of 0.96 (P<0.05) specificity of 1.00, and sensitivity of 0.83, while LR model diagnosed ccRCC with AUC of 0.96 (P<0.05), specificity of 1.00, and sensitivity of 0.83. In the testing set, RF model diagnosed ccRCC with AUC of 0.96 (P<0.05), specificity of 1.00, and sensitivity of 0.89, while LR model diagnosed ccRCC with AUC of 0.88(P<0.05), specificity of 0.90, and sensitivity of 0.77. Conclusion Radiomics models based on enhanced CT can be used for identifying ccRCC from non-ccRCC. RF model has higher diagnostic value than LR model.
查看全文  查看/发表评论  下载PDF阅读器