基于CT特征与影像组学的T1期透明细胞肾细胞癌核分级预测模型构建
回顾性纳入杭州市中医院90例及邵逸夫医院43例T1期ccRCC患者,构建CT常规影像模型、影像组学模型和列线图联合模型预测WHO/ISUP分级。训练集AUC分别为0.742、0.914、0.929,外部验证集为0.664、0.879、0.882,影像组学与联合模型均优于常规影像模型。
要点
- 基于CT皮质期影像组学模型术前预测T1期ccRCC WHO/ISUP分级训练集AUC为0.914。
- 列线图联合模型训练集AUC为0.929,外部验证集为0.882。
- 影像组学与联合模型AUC均高于常规影像模型(均P<0.05)。
核心结论
- 基于CT皮质期影像组学模型及综合常规影像征象的列线图联合诊断模型,术前预测T1期ccRCC WHO/ISUP分级具有良好的效能。
研究设计与样本
- 回顾性收集2016年1月至2023年12月杭州市中医院90例ccRCC患者为训练集,2017年1月至2018年12月浙江大学医学院附属邵逸夫医院43例ccRCC患者为外部验证集。
- 训练集中低级别组64例、高级别组26例;外部验证集中低级别组33例、高级别组10例。
- 训练集基于CT图像特征用多因素logistic回归建立常规影像模型;手动勾画皮质期肿瘤立体感兴趣区提取影像组学特征,经线性相关检查和L1正则化筛选后采用线性支持向量机分类器构建影像组学模型。
- 基于常规影像模型和影像组学评分,通过多因素logistic回归建立列线图联合诊断模型,用ROC曲线评估效能,DeLong检验比较AUC。
主要发现
- 训练集常规影像模型AUC为0.742(95%CI:0.623~0.860),外部验证集为0.664(95%CI:0.448~0.879)。
- 训练集影像组学模型AUC为0.914(95%CI:0.844~0.983),外部验证集为0.879(95%CI:0.718~1.000)。
- 训练集列线图联合诊断模型AUC为0.929(95%CI:0.858~0.999),外部验证集为0.882(95%CI:0.710~1.000)。
- 影像组学模型和列线图联合诊断模型的AUC均高于常规影像模型(均P<0.05)。
- 列线图联合诊断模型与影像组学模型的AUC差异无统计学意义(均P>0.05)。
关键数值与推荐
- 根据WHO/ISUP分级标准,Ⅰ、Ⅱ级为低级别组,Ⅲ、Ⅳ级为高级别组。
临床怎么用
- 原文未据此给出推荐。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
Objective: To investigate the clinical value of the prediction models constructed by CT based imaging features and radiomics for World Health Organization/International Society of Urological Pathology (WHO/ISUP) grading in pre-operative patients with T1 clear cell renal cell carcinoma (ccRCC). Methods: Ninety patients with ccRCC diagnosed at Hangzhou Hospital of Traditional Chinese Medicine from January 2016 to December 2023 were enrolled as the training set, and 43 patients diagnosed at the Sir Run Run Shaw Hospital from January 2017 to December 2018 were enrolled as the external validation set. According to the WHO/ISUP grading system, grades Ⅰ and Ⅱ were defined as the low grade group, and grades Ⅲ and Ⅳ were defined as the high grade group. In the training set, 64 patients were in the low grade group and 26 patients in the high grade group. In the external validation set, 33 patients were in the low grade group and 10 patients in the high grade group. The multivariate logistic regression was used to establish an imaging factor model based on CT imaging features in the training set. The 3-dimensional regions of interest were manually contoured at the cortical phase of enhanced CT, and the radiomics features were extracted. Linear correlation between features and L1 regularization were used for feature selection, and then linear support vector classification was used to construct the radiomics model. After that, a combined diagnostic model of nomogram combining the radiomics score and imaging factors was constructed using multivariate logistic regression analysis. The receiver operating characteristic (ROC) curve was used to evaluate the effectiveness of each model. The Delong test was used for comparison of the areas under the ROC curve. Results: The imaging factor model, the radiomics model, and the combined diagnostic model of nomogram were successfully constructed to predict the WHO/ ISUP grading in stage T1 ccRCC. The AUC value of the imaging factor model in the training and validation sets was 0.742 (95% CI: 0.623-0.860) and 0.664 (95% CI: 0.448-0.879), respectively. The AUC values of the radiomics model in the two sets were 0.914 (95% CI: 0.844-0.983) and 0.879 (95% CI: 0.718-1.000), and of the combined diagnostic model of nomogram in the two sets were 0.929 (95% CI: 0.858-0.999) and 0.882 (95% CI: 0.710-1.000), respectively. The AUCs of the radiomics model and combined diagnostic model of nomogram were significantly higher than that of the imaging factor model (both P<0.05). There was no statistical difference in the AUCs between the combined diagnostic model of nomogram and the radiomics model (both P>0.05). Conclusion: The CT-based radiomics model and combined diagnostic model of nomogram incorporating radiomics signature and imaging features showed favorable predictive efficacy for the preoperative prediction of WHO/ISUP grading in stage T1 ccRCC. 目的: 探讨基于CT征象和影像组学特征构建模型术前预测T1期肾透明细胞癌(ccRCC)WHO/国际泌尿病理学会(ISUP)分级的方法并评价其临床价值。 方法: 回顾性收集2016年1月至2023年12月杭州市中医院收治的90例ccRCC患者为训练集,收集2017年1月至2018年12月浙江大学医学院附属邵逸夫医院收治的43例ccRCC患者为外部验证集。根据WHO/ISUP分级标准,Ⅰ、Ⅱ级为低级别组,Ⅲ、Ⅳ级为高级别组。训练集中低级别组64例,高级别组26例。外部验证集中低级别组33例,高级别组10例。在训练集中,基于ccRCC的CT图像特征,通过多因素logistic回归分析建立常规影像模型。对横轴位CT皮质期图像手动逐层勾画肿瘤立体感兴趣区后提取影像组学特征,通过特征线性相关检查和L1正则化进行特征筛选,采用线性支持向量机分类器构建影像组学模型。基于常规影像模型和影像组学评分,通过多因素logistic回归分析确定T1期ccRCC WHO/ISUP分级的独立影响因素,建立列线图联合诊断模型。采用受试者工作特征曲线分析评估各模型的预测效能,模型间的曲线下面积(AUC)比较采用DeLong检验。 结果: 成功构建预测T1期ccRCC WHO/ISUP分级的常规影像模型、影像组学模型和列线图联合诊断模型。在训练集和外部验证集中,常规影像模型预测T1期ccRCC WHO/ISUP分级的AUC分别为0.742(95% CI:0.623~0.860)和0.664(95% CI:0.448~0.879),影像组学模型的AUC分别为0.914(95% CI:0.844~0.983)和0.879(95% CI:0.718~1.000),列线图联合诊断模型的AUC分别为0.929(95% CI:0.858~0.999)和0.882(95% CI:0.710~1.000),影像组学模型和列线图联合诊断模型的AUC均高于常规影像模型(均P<0.05),列线图联合诊断模型与影像组学模型的AUC差异无统计学意义(均P>0.05)。 结论: 基于CT皮质期的影像组学模型,以及在此基础上综合常规影像征象的列线图联合诊断模型,术前预测T1期ccRCC WHO/ISUP分级具有良好的效能。.
引用
- Zhao CY, Chen C, Li WW, et al. [Establishment of nuclear grade prediction model for T1 clear cell renal cell carcinoma based on CT features and radiomics]. Zhonghua Zhong Liu Za Zhi. 2025. PMID: 39939018. doi:10.3760/cma.j.cn112152-20240615-00257
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。