基于T2WI影像组学及联合诊断模型鉴别腮腺多形性腺瘤与腺淋巴瘤  

Radiomic analysis based on MR-T2WI and combined diagnostic model for identification of pleomorphic adenoma from adenolymphoma of parotid gland

作　　者：吴艳[1] 谢元亮[1] 张树桐 王翅鹏 马锋 王翔[1] WU Yan;XIE Yuan-liang;ZHANG Shu-tong(Department of Radiology, Wuhan Central Hospital,Tongji Medical College,Huangzhong University of Science and Technology,Wuhan 430014,China)

机构地区：[1]华中科技大学同济医学院武汉中心医院放射科,武汉430014 [2]武汉平安好医医学影像诊断中心

出　　处：《放射学实践》2020年第12期1525-1531,共7页Radiologic Practice

摘　　要：目的:探讨基于磁共振T2WI的影像组学模型对腮腺多形性腺瘤与腺淋巴瘤的鉴别诊断价值。方法:回顾性分析2015年1月-2019年11月在本院经手术病理证实且具有完整术前MR平扫图像的99例腮腺肿瘤患者的病例资料,将患者按7:3的比例随机分为2组:训练集70例,验证集29例。按照术后病理结果,将训练集和验证集均进一步分为多形性腺瘤组和腺淋巴瘤组。使用MaZda软件进行纹理分析,在每例患者T2WI上于肿瘤最大层面勾画ROI,提取310个纹理特征;采用R语言软件对纹理数据进行预处理,并采最小冗余最大相关(mRMR)算法对每例患者提取的310个纹理特征进行降维;然后采用Lasso回归分析及10折交叉验证法进一步筛选纹理特征,用以建立影像组学标签。基于建立的影像组学标签及患者的临床资料,采用多变量Logistic回归分析建立联合诊断模型。采用ROC曲线评估影像组学标签及联合诊断模型的诊断效能。采用Hosmer-Lemesow拟合优度检验分析诊断模型的拟合度。结果:通过降维、筛选后最终保留8个纹理特征,建立的影像组学标签(RS)的计算公式为RS=0.251×Vertl_GLevNonU+0.134×Skewness+0.227×S(5,5)Correlat+0.408×X45dgr_LngREmph-0.131×Teta4+0.187×WavEnHH_s.3+0.183×S(5,-5)Correlat-0.027×Teta1+0.201。在训练集和验证集中,影像组学标签鉴别腮腺多形性腺瘤与腺淋巴瘤的AUC分别为0.83(95%CI:0.73~0.93)和0.82(95%CI:0.64~1.00)。基于多变量Logistic回归分析,最终将性别、单发或多发、病灶位置及影像组学标签作为独立的影响因子纳入联合诊断模型,这4项指标的优势比(OR)分别为0.177(95%CI:0.027~0.878)、15.608(95%CI:1.090~736.275)、4.876(95%CI:3.768~10.754)和9.729(95%CI:2.644~50.430)。训练集和验证集中,联合诊断模型鉴别2类肿瘤的AUC分别为0.90(95%CI:0.83~0.97)和0.96(95%CI:0.88~1.00),均高于影像组学标签。Hosmer-Lemesow拟合优度检验结果显示,在训练集和验证集中,模�Objective:To explore the value of MR T2WI-based radiomics in differentiation diagnosis of pleomorphic adenoma and adenolymphoma of parotid gland.Methods:In this retrospective study,the data of 99 patients with histologically confirmed pleomorphic adenoma or adenolymphoma of parotid gland were collected from January 2015 to November 2019 in our hospital.The primary dataset were divided into training set(70 cases)and verification set(29 cases)according to the ratio of 7:3,and then the training set and verification set were further divided into pleomorphic adenoma group and adenolymphoma group respectively,according to the postoperative pathological results.Mazda software was used to delineate the ROI along the largest tumor boundary in T2WI of each patient and 310 texture features wad extracted;R-language software was used to preprocess the data;mimimum-redundancy maximum-relevancy(mRMR)algorithm was used to data dimension reduction;then Lasso regression analysis and ten-fold cross validation were used to further filter texture features,and finally the texture features filtered and retained were used to establish radiomics signature(RS).The radiomics signature and clinical data of the patients was analyzed using multivariate logistic regression to establish a combined diagnostic model.ROC curve was used to evaluate the diagnostic efficacy of the radiomics signature and the combined diagnostic model.Hosmer lemesow test was used for the goodness of fit of the model.Results:The radiomics signature consisted of 8 selected features,and its formula was as follows:0.251×Vertl_GLevNonU+0.134×Skewness+0.227×S(5,5)Correlat+0.408×X45dgr_LngREmph-0.131×Teta4+0.187×WavEnHH_s.3+0.183×S(5,-5)Correlat-0.027×Teta1+0.201.In the training set and verification set,the AUC of the radiomics signature for differen-tiation pleomorphic adenoma from adenolymphoma was 0.83(95%CI:0.73~0.93)and 0.82(95%CI:0.64~1.00),respectively.By multivariate logistic regression analysis,four index(sex,single or multiple,position and radiomics signatu

关 键 词：腮腺肿瘤 多形性腺瘤 腺淋巴瘤 磁共振成像 影像组学 联合诊断模型 

分 类 号：R445.2[医药卫生—影像医学与核医学] R739.87[医药卫生—诊断学]