基于CT平扫纹理分析的影像组学预测肾透明细胞癌WHO/ISUP分级的初步研究

目的探讨基于肾脏CT平扫图像纹理分析的影像组学模型在预测肾透明细胞癌(ccRCC)WHO/国际泌尿病理学会(ISUP)病理分级中的价值。方法回顾性分析2016年12月至2019年5月中国科学院大学附属肿瘤医院经手术病理证实且有明确病理分级的90例ccRCC患者,按照7∶3的比例将所有患者随机分为训练组(63例)及测试组(27例)。根据2016版WHO/ISUP分级标准,将Ⅰ、Ⅱ级归为低级别组(53例),Ⅲ、Ⅳ级归为高级别组(37例)。在CT平扫图像上逐层勾画肿瘤ROI,提取93个纹理特征,利用最小绝对收缩与选择算子(LASSO)回归对特征参数进行降维,并建立影像组学评分(Rad-score)。以病理分级结果为金标准,采用logistic回归构建ccRCC病理分级的预测模型。采用ROC曲线及校准曲线评价模型的诊断效能,计算曲线下面积(AUC)、敏感度、特异度和准确度。采用Hosmer-Lemeshow拟合优度检验评价模型的校准度。结果经降维和交叉验证后筛选出10个非零系数的纹理特征,根据这10个特征及其对应系数的线性加权形成预测ccRCC新病理分级的影像组学风险评分,并建立预测模型。该模型在训练组中的AUC值为0.933(95%CI 0.862~1.000),其判断WHO/ISUP分级高级别ccRCC的灵敏度为92.3%,特异度为89.2%,准确度为90.5%,校准曲线显示该模型的校准度较好(P=0.257)。在测试组中的AUC值为0.875(95%CI 0.734~1.000),灵敏度为72.7%,特异度为87.5%,准确度为81.5%,校准曲线显示该模型的校准度较好(P=0.125)。结论基于平扫CT纹理分析构建的影像组学预测模型对ccRCC WHO/ISUP病理分级的评估具有应用潜能。

A preliminary study of radiomics in predicting WHO/ISUP grading of clear cell renal cell carcinoma based on unenhanced CT texture analysis

Objective To investigate the value of radiomics based on unenhanced CT texture analysis in predicting the WHO/International Society of Urological Pathology(ISUP)grading of clear cell renal cell carcinoma(ccRCC).Methods Postoperative pathology-confirmed ccRCC subjects(n=90)who received CT scanning and had a definite pathological grading in Cancer Hospital of the University of Chinese Academy of Sciences were collected retrospectively from December 2016 to May 2019.The cases were randomly divided into training group(n=63)and test group(n=27)as a ratio of 7∶3.All cases were classified into low grade(gradesⅠandⅡ,n=57)and high grade(gradesⅢandⅣ,n=37)according to the new pathological grading(WHO/ISUP grading,version 2016)of renal carcinoma.3D-ROI segmentation was performed on unenhanced CT images and 93 texture features were extracted.The least absolute shrinkage and selection operator(LASSO)regression was used to reduct dimension of texture parameters and then the radiomics score(Rad-score)was established.The logistic regression was used to develop the prediction model with the pathological grading as the gold standard.The ROC curve and calibration curve were used to evaluate the predictive performance of the model,and the area under the curve(AUC),accuracy,sensitivity and specificity were calculated.The Hosmer-Lemeshow test was used to evaluate calibration degree of the model.Results The 10 non-zero coefficient texture features were screened out through dimension reduction steps.The Rad-score was formed according to the linear combination of these ten features and corresponding coefficients,and then the prediction model was developed.The AUC of the model in training group was 0.933(95%CI 0.862-1.000),the sensitivity was 92.3%,the specificity was 89.2%,and the model accuracy was 90.5%.The calibration curve showed the good calibration(P=0.257).The AUC value in test group was 0.875(95%CI 0.734-1.000),the sensitivity,specificity and accuracy were 72.7%,87.5%and 81.5%.The calibration curve showed the good calibration(P=0.125).Conclusion The radiomics prediction model based on unenhanced CT texture analysis have application potential for the evaluation of WHO/ISUP grading of ccRCC.