基于CT皮质期影像组学预测肾细胞癌亚型的研究
A study of radiomics for predicting renal cell carcinoma subtypes based on cortical phase CT image
赵才勇 1陈文 1严志强 1陈超 2崔凤1
作者信息
- 1. 杭州市中医院放射科,浙江 杭州 310007
- 2. 浙江大学医学院附属邵逸夫医院放射科,浙江 杭州 310016
- 折叠
摘要
目的:探讨基于CT皮质期影像组学鉴别肾透明细胞癌(ccRCC)和非透明细胞癌(non-ccRCC)的价值.方法:回顾性分析 2017 年 1 月—2022 年 12 月经病理证实的 122 例肾细胞癌患者的资料,其中ccRCC 82 例,non-ccRCC 40 例,并以随机数表法按 7∶3 的比例将患者分成训练集(n=85)和验证集(n=37).在CT皮质期手工逐层勾画肿瘤感兴趣区(ROI)后提取影像组学特征,使用特征间线性相关检查和F检验依次进行特征筛选,采用逻辑回归分类器构建影像组学模型.采用t检验、χ2 检验及Logistic回归分析筛选CT影像特征,建立常规影像模型.综合影像组学评分和常规影像模型建立联合模型.绘制ROC曲线评估各模型的预测效能,AUC比较采用Delong检验.结果:影像组学模型在训练集和验证集中的AUC分别为 0.990(95%CI 0.976～1.0)和 0.890(95%CI 0.774～1.0).在训练集和验证集中,影像组学模型和联合模型的预测效能均优于常规影像模型,差异有统计学意义(P均<0.05);相比联合模型,在验证集中影像组学模型的预测效能略高,但无统计学差异(P=0.27).结论:基于CT皮质期影像组学模型对预测肾细胞癌亚型具有较好的效能.
Abstract
Objective:To investigate the value of radiomics based on cortical phase CT image in distinguishing clear cell renal cell carcinoma(ccRCC)and non-clear cell renal cell carcinoma(non-ccRCC).Methods:A total of 122 patients diagnosed as ccRCC(n=82)or non-ccRCC(n=40)by pathology from January 2017 to December 2022 were retrospectively analyzed.Pa-tients were randomly assigned to a training cohort(n=85)and a validation cohort(n=37)in a ratio of 7∶3.The 3-dimensional regions of interest(ROIs)were manually contoured at the cortical phase,and the radiomics features were extracted.Linear cor-relation between features and F-test were used for feature selection and then Logistic regression was used to construct the ra-diomics model.CT imaging features were selected using t-test,χ2-test and Logistic regression to build a conventional imaging model.A joint model was established by combining the radiomic model and conventional imaging model.Receiver operating characteristic(ROC)curves were plotted to evaluate the predictive performance of each model.Delong test was used for com-parison of AUC values between every two models.Results:The AUC values of the radiomics model were 0.99(95%CI 0.976～1.0)and 0.89(95%CI 0.774～1.0)in the training and validation cohort,respectively.In the training and validation cohorts,the predictive performance of radiomics model and joint model were superior to the conventional imaging model,with statistically significant differences(all P<0.05).Compared with the joint model,the predictive performance of the radiomics model was slightly higher in the validation cohort,but there was no statistical difference(P=0.27).Conclusions:The radiomics model based on cortical phase CT image showed favorable predictive efficacy for predicting renal cell carcinoma subtypes.
关键词
癌,肾细胞/体层摄影术,螺旋计算机Key words
Carcinoma,Renal Cell/Tomography,Spiral Computed引用本文复制引用
基金项目
杭州市生物医药和健康产业发展扶持科技项目(2021WJCY355)
出版年
2024
国家科技期刊平台
NSTL
万方数据