基于T1WI增强影像组学预测WHOⅡ-Ⅲ级脑胶质瘤Ki-67指数表达的研究  

Contrast enhanced T1WI-based radiomics for preoperatively predicting expression status of Ki-67 proliferation index in WHO gradeⅡ-Ⅲgliomas

作　　者：杨瀚 樊建坤 陈光祥 唐光才 周鹏[1] 赵娟 YANG Han;FAN Jiankun;CHEN Guangxiang;TANG Guangcai;ZHOU Peng;ZHAO Juan(Department of Radiology,Sichuan Cancer Hospital&Institute(i.e.Sichuan Cancer Center,or Cancer Hospital Afiliated to University of Electronic Science and Technology of China),Sichuan Chengdu 610041,China;Department of Radiology,Jianyang People's Hospital,Sichuan Chengdu 641400,China;Department of Radiology,the Afiliated Hospital of Southwest Medical University,Sichuan Luzhou 646000,China;Department of Anesthesiology,Sichuan Cancer Hospital&Institute(i.e.Sichuan Cancer Center,or Cancer Hospital Afiliated to University of Electronic Science and Technology of China),Sichuan Chengdu 610041,China)

机构地区：[1]四川省肿瘤医院·研究所四川省癌症防治中心电子科技大学附属肿瘤医院影像科,四川成都610041 [2]简阳市人民医院放射科,四川成都641400 [3]西南医科大学附属医院放射科,四川泸州646000 [4]四川省肿瘤医院·研究所四川省癌症防治中心电子科技大学附属肿瘤医院麻醉科,四川成都610041

出　　处：《现代肿瘤医学》2024年第12期2244-2249,共6页Journal of Modern Oncology

基　　金：国家自然科学基金资助项目(编号:U21A20521)。

摘　　要：目的:探讨基于术前T1WI增强图像建立的影像组学模型预测WHO II-III脑胶质瘤Ki-67表达水平的可行性及价值。方法:纳入经我院术后病理及免疫组化确诊的WHO II-III脑胶质瘤119例,其Ki-67高表达组51例,Ki-67低表达组68例,按8∶2随机分为训练组(n=96)和验证组(n=23)。使用ITK-SNAP软件在轴位T1WI增强序列上逐层手动勾画病灶所有层面的感兴趣区,包含瘤体及瘤周水肿。使用Python软件内Radiomics模块提取三维影像组学特征并进行Z-score标准化,应用组内相关系数(intra-class correlation coefficient,ICC)、Spearman相关系数和最小绝对收缩和选择算法(least absolute shrinkage and selection operator,LASSO)筛选出5个最优影像组学特征并建立逻辑回归模型。采用受试者工作特征曲线对模型进行效能评价。结果:影像组学预测模型在训练组曲线下面积为0.926(95%CI:0.861~0.978),敏感度为90.2%,特异度为81.9%,准确度为85.7%,阳性预测值为81.2%,阴性预测值为88.3%;在验证组中曲线下面积为0.801(95%CI:0.725~0.985),敏感度为85.2%,特异度为78.6%,准确度为79.9%,阳性预测值为79.3%,阴性预测值为82.9%。结论:基于术前T1WI增强图像建立的影像组学模型能够有效预测WHOⅡ-Ⅲ级脑胶质瘤Ki-67表达水平。Objective:To explore the the feasibility and value of a radiomics model based on preoperative T1WI-enhanced imaging in predicting Ki-67 proliferation index(PI)expression level in glioma patients with WHO gradeⅡ-Ⅲ.Methods:A total of 119 patients with WHO gradeⅡ-Ⅲgliomas(51 cases with Ki-67 PI≥10%and 68 cases with Ki-67 PI<10%)who were confirmed by pathology and immunohistochemistry were retrospectively recruited,and randomly divided into a training dataset(n=96)and a validation dataset(n=23)at the ratio of 8∶2.3D region of interest was manually delineated slice-by-slice on the axial plane for T1WI-enhanced images covering the tumor and peripheral edema area via the ITK-SNAP software.Radiomics module in Python software was used to extract and standardise the features,and five radiomics features were finally selected by intra-class correlation coefficient(ICC)and Spearman correlation coefficient.A Logistic regression prediction model was finally constructed via least absolute shrinkage and selection operator(LASSO).Receiver operating characteristic curve analysis was performed to evaluate the performance of the model.Results:The radiomics model showed good performance in predicting the Ki-67 PI expression level in both the training dataset(area under curve 0.926,95%CI:0.861~0.978,sensitivity 90.2%,specificity 81.9%,accuracy 85.7%,positive predictive value 81.2%,negative predictive value 88.3%,respectively)and the validation dataset(area under curve 0.801,95%CI:0.725~0.985,sensitivity 85.2%,specificity 78.6%,accuracy 79.9%,positive predictive value 79.3%,negative predictive value 82.9%,respectively).Conclusion:The radiomics model based on the preoperative T1WI-enhanced images can effectively predict the Ki-67 expression level in patients with WHO gradeⅡ-Ⅲ gliomas.

关 键 词：影像组学 磁共振成像 脑胶质瘤 KI-67 

分 类 号：R739.41[医药卫生—肿瘤]