基于MRI-T2WI影像组学列线图鉴别卵巢交界性与恶性上皮源性肿瘤  

Differentiation of borderline and malignant epithelial tumors based on MRI-T2WI radiomics nomogram

作　　者：丁聪 魏明翔 贾建业 周围 柏根基[1] DING Cong;WEI Mingxiang;JIA jianye;ZHOU Wei;BAI Genji(Department of Imaging,the Affiliated Huaian No.1 People's Hospital of Nanjing Medical University,Huaian 223000,China;Department of Imaging,the Affiliated Suzhou Hospital of Nanjing Medical University,Suzhou 215000,China)

机构地区：[1]南京医科大学附属淮安第一医院影像科,淮安223000 [2]南京医科大学附属苏州医院影像科,苏州215000

出　　处：《磁共振成像》2022年第7期55-60,共6页Chinese Journal of Magnetic Resonance Imaging

基　　金：北京医卫健康公益基金(编号:B20240ES)。

摘　　要：目的建立基于MRI-T2WI影像组学列线图并评价其鉴别卵巢交界性上皮源性肿瘤(boderline epithelial ovarian tumors,BEOTs)及恶性上皮源性肿瘤(malignant epithelial ovarian tumors,MEOTs)的效能及临床应用价值。材料与方法回顾性分析2016年1月至2021年5月间南京医科大学附属淮安第一医院经病理证实的上皮源性卵巢肿瘤患者的临床及影像资料,共计192例,其中BEOTs 72例,MEOTs 120例,按8∶2比例随机分为训练集(n=153)及测试集(n=39),从每个患者的轴位T2WI图像中手动勾画感兴趣区并提取影像组学特征。使用Mann-Whitney U检验、相关性分析及最小绝对收缩选择算子(least absolute shrinkage and selection operator,LASSO)回归进行特征选择,并构建影像组学模型及计算影像组学评分Radscore。结合临床因素及Radscore,采用多元logistic回归模型构建影像组学列线图模型。最后通过ROC曲线、校准曲线及决策曲线分析评价列线图模型的临床应用价值。结果经特征筛选后最终保留10个影像组学特征。结合HE4和Radscore的影像组学列线图在训练集及测试集中的曲线下面积值(area under the cure,AUC)(训练集:0.947,测试集:0.914)均大于单一的影像组学模型(训练集:0.925,测试集:0.819)。ROC曲线及决策曲线分析结果显示,影像组学列线图模型更具优势。结论结合MRI-T2WI影像组学特征和临床因素的影像组学列线图模型可直观、准确地鉴别BEOTs及MEOTs,并为下一步的临床决策提供指导。Objective:To develop and validate a radiomics nomogram that was based on MRI-T2WI to distinguish between borderline epithelial ovarian tumors(BEOTs)and malignant epithelial ovarian tumors(MEOTs).Materials and Methods:The clinical and imaging data of 192 patients with epithelial ovarian tumors confirmed by pathology from January 2016 to May 2021 were retrospectively analyzed in the Affiliated Huaian First People's Hospital of Nanjing Medical University,including EBOTs(n=72)and MEOTs(n=153)were enrolled.According to the ratio of 8∶2,all cases were randomly divided into the training group(n=153)and validation group(n=39).We used T2WI to manually delineated ROI and extract radiomics features.Mann-Whitney U test,correlation and LASSO regression were used to select features,and then constructed radiomics model by these features,used to calculate Radscore.Combining Radscore with clinic factors,we used multiple logistic regression to construct radiomics nomogram.ROC curve,calibration curve and decision curve analysis and correction were used to evaluate the clinical value of radiomics nomogram.Results:We reserved 10 radiomics features after the feature was filtered.The AUC of the radiomics nomogram which combined HE4 with Radscore in the training group and validation group(training group:0.947,validation group:0.914)were higher than those of the single radiomics model(training group:0.925,validation group:0.819).ROC and DCA results showed that the radiomics nomogram had higher reliability.Conclusions:The radiomics nomogram combined radiomics feature based on T2WI and clinical factors is able to distinguish between BEOTs and MEOTs intuitively and accurately and provide guidance for the next clinical decision.

关 键 词：卵巢肿瘤 影像组学 机器学习 列线图 磁共振成像 T2加权成像 

分 类 号：R445.2[医药卫生—影像医学与核医学] R737.31[医药卫生—诊断学]