MRI影像组学对乳头状肾细胞癌WHO/ISUP分级的预测研究  

Prediction of WHO/ISUP Grades in Papillary Renal Cell Carcinoma Using MRI Radiomics

作　　者：赵厚铭 贾通宇 于淼[3] 白旭 王集琛 许清江 李尚韦 宋佳龙 杨国强[2] 丁效蕙[3] 黄庆波[2] 马鑫[2] Zhao Houming;Jia Tongyu;Yu Miao;Bai Xu;Wang Jichen;Xu Qingjiang;Li Shangwei;Song Jialong;Yang Guoqiang;Ding Xiaohui;Huang Qingbo;Ma Xin(Medical School of Chinese PLA,Beijing 100853,China;Department of Urology,the Third Medical Center of PLA General Hospital,Beijing 100039,China;Department of Pathology,the First Medical Center of PLA General Hospital,Beijing 100853,China;Department of Radiology,the First Medical Center of PLA General Hospital,Beijing 100853,China)

机构地区：[1]解放军医学院,北京100853 [2]解放军总医院第三医学中心泌尿外科医学部,北京100039 [3]解放军总医院第一医学中心病理科,北京100853 [4]解放军总医院第一医学中心影像科,北京100853

出　　处：《微创泌尿外科杂志》2025年第1期6-13,共8页Journal of Minimally Invasive Urology

基　　金：国家重点研发计划(2023YFB4706000)。

摘　　要：目的:探讨基于机器学习的磁共振成像(MRI)影像组学模型术前预测乳头状肾细胞癌(pRCC)世界卫生组织/国际泌尿病理协会(WHO/ISUP)分级的临床价值。方法:回顾性收集2010年1月至2023年12月于解放军总医院第一医学中心行手术治疗术后病理证实为pRCC的153例患者,采用WHO/ISUP病理分级标准将其分为低级别组(Ⅰ和Ⅱ级)和高级别组(Ⅲ和Ⅳ级)。通过单因素和多因素分析确定临床独立预测指标。提取MRI图像上的影像组学特征,采用最小冗余最大相关性(mRMR)、最小绝对收缩和选择算子(LASSO)回归等方式筛选影像组学特征,使用使用Logistic回归和支持向量机(SVM)构建临床模型、影像组学模型以及临床-影像组学联合模型。使用受试者工作特征(ROC)曲线下面积(AUC)评估3个模型预测效果,DeLong检验比较3个模型的AUC值,校准曲线评估模型的校准度,决策曲线(DCA)评价临床效能。结果:4个MRI序列共提取到3776个影像组学特征,筛选后最终获得13个特征用于构建模型。多因素分析确定临床指标中肿瘤最大径和全身炎症反应指数(SIRI)是WHO/ISUP分级的独立预测指标。训练集中,影像组学模型的表现(AUC=0.837)优于临床模型(AUC=0.776),临床-影像组学联合模型预测效能最佳(AUC=0.889),并经DeLong检验显示显著优于临床模型(P=0.017)。验证集中,联合模型(AUC=0.853)的表现显著优于临床模型(AUC=0.725,P=0.019)和影像组学模型(AUC=0.826,P=0.106)。校准曲线显示,联合模型的WHO/ISUP分级预测在训练集和验证集中与实际结果接近。DCA显示,与影像组学模型和临床模型相比,联合模型具有更高的净效益。结论:基于机器学习的多序列MRI的影像组学模型是一种有效的无创检测工具,在术前预测pRCC WHO/ISUP分级方面显示出良好的效果。Objective:To investigate the clinical utility of a machine learning(ML)-based magnetic resonance imaging(MRI)radiomics approach for the preoperative prediction of World Health Organization/International Society of Urological Pathology(WHO/ISUP)grades in papillary renal cell carcinoma(pRCC).Methods:A retrospective collection of 153 pRCC patients,confirmed by postoperative pathology after surgical treatment at the First Medical Center of the PLA General Hospital from January 2010 to December 2023,was performed.According to the WHO/ISUP pathological grading standard,patients were classified into low-grade group(grades I and II)and high-grade group(grades III and IV).Univariate and multivariate analyses were conducted to determine independent clinical pre-dictive factors.MRI images were analyzed for radiomic features,which were selected using methods such as mini-mum redundancy maximum relevance(mRMR)and least absolute shrinkage and selection operator(LASSO)regres-sion.Clinical,radiomic,and clinical-radiomic combined models were constructed using Logistic regression and a sup-port vector machine(SVM).The predictive performance of the three models was evaluated using receiver operating characteristic(ROC)curves and the area under the curve(AUC).DeLong's test was used to compare the AUC val-ues of the models,and calibration curves were employed to assess model calibration.Decision curve analysis(DCA)was used to evaluate clinical utility.Results:A total of 3,776 radiomic features were extracted from four MRI se-quences,and after selection,13 features were used to build the models.Multivariate analysis identified the maximum tumor diameter and the systemic inflammation response index(SIRI)as independent predictors of WHO/ISUP grad-ing.In the training set,the radiomic model(AUC=0.837)outperformed the clinical model(AUC=0.776),and the clinical-radiomic combined model demonstrated the best predictive performance(AUC=0.889),significantly better than the clinical model(P=0.017)according to DeLong's test.In the validation set,

关 键 词：乳头状肾细胞癌 影像组学 世界卫生组织/国际泌尿病理协会分级 预测模型 

分 类 号：R737.11[医药卫生—肿瘤]