基于CT影像组学预测肾透明细胞癌ISUP分级及病理分期的研究  

Radiomics Model for Predicting the Pathological Grade and Stage of Clear Cell Renal Cell Carcinoma Based on CT Imaging

作　　者：王煦[1] 洪楠[1] 尹平 孙超[1] WANG Xu;HONG Nan;YIN Ping(Department of Radiology,Peking University People’s Hospital,Beijing 100044,P.R.China)

机构地区：[1]北京大学人民医院放射科,100044

出　　处：《临床放射学杂志》2022年第4期670-675,共6页Journal of Clinical Radiology

基　　金：国家重点研发计划项目(子课题)(编号:2017YFC0109003)。

摘　　要：目的 探讨基于CT影像组学方法术前预测肾透明细胞癌(ccRCC)的病理分级和分期的可行性。方法 搜集本院2015年1月至2018年6月泌尿外科收治的125例经病理证实的ccRCC患者。根据肾细胞癌(RCC)世界卫生组织(WHO)/国际泌尿病理协会(ISUP)(2012)组织学分级系统分为低级别组(核Ⅰ~Ⅱ级)66例和高级别组(核Ⅱ~Ⅲ、Ⅲ级和Ⅳ级)59例;之后随机按照7∶3的比例分入训练组和测试组,选取出最具预测意义的特征,并采用随机森林法建立影像组学模型,最终使用受试者工作特征(ROC)曲线和准确率评估模型的诊断效能。再根据病理分期,将患者分为低分期组65例和高分期组60例,也建立相应影像组学模型并测试预测效能。结果 共5个特征被筛选入鉴别核低级别和核高级别的模型中。该模型在训练组曲线下面积(AUC)=0.946,灵敏度和特异度分别为0.891、0.87,阳性预测值和阴性预测值分别为0.872和0.889,准确率为0.88;在测试组,AUC=0.876,灵敏度和特异度分别为0.75和0.7,阳性预测值和阴性预测值分别为0.872和0.879,准确率为0.725。共筛选出7个特征用于构建鉴别病理低分期组和高分期组的模型中,其在训练组AUC=0.974,灵敏度和特异度分别为0.911、0.911,阳性预测值和阴性预测值分别为0.911和0.911,准确率为0.911;测试组AUC=0.751,灵敏度和特异度分别为0.8和0.45,阳性预测值和阴性预测值分别为0.593和0.692,准确率为0.625。结论 基于CT的影像组学模型对术前预测ccRCC细胞核分级有着较好的效果,而预测病理分期效果稍差。Objective To explore the feasibility of preoperatively predicting the pathological grade and staging of clearcell renal cell carcinoma(ccRCC) using radiomics models based on CT-enhanced imaging.Methods 125 cases of patho-logically confirmed ccRCC patients admitted to the Department of Urology in our hospital from 2015 to 2018 were collectedretrospectively.According to the RCC WHO/ISUP(2012) histological grading system,66 cases were divided into the low-level group(nuclear Ⅰ-Ⅱ) and 59 cases in the high-level group(nuclear Ⅱ-Ⅲ,Ⅲ and Ⅳ);Then,all cases were randomly divided into the training group and testing group according to the radio of 7:3.The most predictive features wereselected and the random forest method was used to establish the radiomics model.Finally,the ROC curve and accuracy wereused to evaluate the diagnostic efficacy of the model.According to the pathological stage,65 patients were divided into low-stage group and 60 cases were divided into high-stage group.The corresponding radiomics model was also established andthe predictive power was validated.Results A total of 5 features were selected into the model to distinguish low-level nu-clear from high-level nuclear.An AUC = 0.946 in the training group was achieved,the sensitivity and specificity were 0.891and 0.87,respectively.The positive prediction rate and the negative prediction rate were 0.872 and 0.889,and the accuracywas 0.88.In the testing group,an AUC = 0.876 was obtained.The sensitivity and specificity were respectively 0.75 and 0.7,the positive prediction rate and the negative prediction rate were 0.872 and 0.879,respectively,and the accuracy was 0.725.A total of 7 features were screened to construct a model to distinguish low-stage from high-stage pathological groups.Inthe training group,the AUC = 0.974,the sensitivity and specificity were 0.911 and 0.911,respectively.The positive predic-tion rate and negative prediction rate were 0.911 and 0.911,respectively,and the accuracy is 0.911.In the testing group,an AUC = 0.751 was achi

关 键 词：肾透明细胞癌 影像组学 评估模型 病理分级 分期 

分 类 号：R737.11[医药卫生—肿瘤] R730.44[医药卫生—临床医学]