基于VHL基因表达构建CT组学特征预测ccRCC Fuhrman分级的诊断模型  

CT Omics Based on Von Hippel-Lindau Gene Expression to Predict the Diagnosis of Fuhrman Grading of ccRCC

作　　者：田晋捷 马新伟 许建铭 庞洪权 朱建兵 TIAN Jinjie;MA Xinwei;XU Jianming;PANG Hongquan;ZHU Jianbing(Department of Radiology,Suzhou Science&Technology Town Hospital,Suzhou Jiangsu 215000,China)

机构地区：[1]苏州科技城医院影像科,江苏苏州215000

出　　处：《中国医疗设备》2022年第8期118-122,共5页China Medical Devices

基　　金：苏州市科技局-关键技术攻关项目(SKY2021031);卫生计生委科技项目临床重点病种诊疗技术专项(LCZX201930);苏州市卫生科技计划青年临床应用研究项目(SKJY2021035)。

摘　　要：目的探讨Von Hippel-Lindau(VHL)基因表达与肾透明细胞癌(Clear Cell Renal Cell Carcinoma,ccRCC)Fuhrman分级之间的关系,并根据VHL表达匹配相关CT组学特征,构建CT组学特征预测ccRCC Fuhrman分级的诊断模型。方法本研究所有病例均来自TCGA数据库,共163例经病理证实的ccRCC病例。在入组病例的CT图像上构建瘤体体积(Tumor Mass Volume,TMV),每例ccRCC癌灶中计算588个影像组学特征。将病例分成训练组(n=120)与测试组(n=43),在训练组中,通过秩相关分析选择与ccRCC Fuhrman分级具有显著统计学差异的CT组学特征(P<0.01),并用筛选出的有统计学差异的组学特征对ccRCC Fuhrman分级进行受试者工作特征(Receiver Operator Characteristic,ROC)曲线诊断效能检验,初步构建以组学特征和学习分类器为主要组成的诊断模型。在测试组中,采用训练组筛选出的有统计学差异的组学特征对ccRCC Fuhrman分级进行ROC诊断效能测试,验证诊断模型的预测效能,构建有效CT组学特征预测ccRCC Fuhrman分级的诊断模型。结果VHL表达量在ccRCC Fuhrman高低级别组中分布存在统计学差异(P=0.037)。在训练组中通过相关性分析,筛选出25个有统计学差异的CT组学特征,对ccRCC Fuhrman高低分级具有较高的诊断效能,曲线下面积(Area Under the Curve,AUC)(95%CI)为0.742(0.654~0.817),敏感度79.0%,特异度61.4%,构建预测诊断模型。在测试组中,对模型的诊断效能进行验证,25个CT组学特征对ccRCC Fuhrman高低分级诊断效能AUC值为0.816,95%CI:0.668~0.918,敏感度90.9%,特异度61.9%。结论VHL基因突变相关的CT组学特征模型对ccRCC Fuhrman高低分级预测具有较高的诊断效能。Objective To investigate the relationship between Von Hippel-Lindau(VHL)gene expression and Fuhrman grade of clear cell carcinoma of kidney(ccRCC),and to construct a diagnostic model for predicting Fuhrman grade of ccRCC based on CT omics characteristics matching VHL expression.Methods All the cases in this study were from TCGA database,including 163 pathologically confirmed cases of ccRCC.Tumor mass volume(TMV)was constructed on CT images of the enrolled cases.A total of 588 imaging features were calculated in each ccRCC cancer focus.Cases were divided into the training group(n=120)and the test group(n=43).In the training group,rank correlation analysis was used to select CT histological features that were statistically different from ccRCC Fuhrman grading(P<0.01).The ccRCC Fuhrman grade was tested by receiver operator characteristic(ROC)diagnostic efficiency using the selected omics features with statistical difference.A diagnostic model consisting of omics features and learning classifiers was constructed.In the test group,the ccRCC Fuhrman grade was tested by ROC diagnostic efficiency using the statistically significant omics features screened out by the training group,and the diagnostic model of effective CT omics features to predict ccRCC Fuhrman grade was constructed.Results The distribution of VHL expression in ccRCC Fuhrman grade groups was statistically different(P=0.037).Through correlation analysis,25 CT omics features with statistical differences were screened out in the training group,which had high diagnostic efficacy for ccRCC Fuhrman classification,with AUC value of 0.742(0.654~0.817),sensitivity of 79.0%,specificity of 61.4%.The construct predictive diagnostic model was built.In the test group,to verify the diagnostic efficacy of the model,the AUC value of 25 CT omics features for high-low grade Fuhrman diagnosis of ccRCC was 0.816,95%CI:0.668~0.918,sensitivity 90.9%,specificity 61.9%.Conclusion The CT omics model related to VHL gene mutation has high diagnostic efficacy for Fuhrman classificati

关 键 词：VHL基因 CT组学特征 肾透明细胞癌 预测模型 

分 类 号：R814.4[医药卫生—影像医学与核医学]