基于磁共振成像的影像组学特征在脑胶质瘤分级中的应用

Application value of radiomics features based on magnetic resonance imaging in grading of cerebral of gliomas

目的基于MRI T2WI序列构建的影像组学模型区分脑胶质瘤Ⅱ级和Ⅲ级的诊断效能。方法本研究影像学数据均来自于癌症基因组数据集,收集了梅奥诊所于2002年10月~2011年8月进行MRI平扫及肿瘤分级的159例大脑胶质瘤数据,包括胶质Ⅱ级(n=104)和Ⅲ级(n=55),按7:3的比例将患者分为训练组(n=111)和验证组(n=48)。将T2WI图像导入ITK-SNAPv.3.4.0软件,手动勾画大脑全肿瘤感兴趣区(VOI),然后将勾画过的数据导入A.K软件用于提取肿瘤的影像组学特征,共396个特征被提取,主要特征包括6类,分别为直方图、灰度共生矩阵、灰度大小区域矩阵、灰度行程矩阵、形状及Haralick。采用LASSO回归模型进行进一步的特征筛选。根据筛选得到的特征与其相应回归系数的加权线性组合构建影像组学模型,并据此计算每位患者的影像组学评分。在训练组和验证组中采用ROC曲线下面积(AUC)评估影像组学模型的预测性能。影像组学的校准度采用Hosmer-Lemeshow检验进行评价。通过采用决策曲线分析进行评估影像组学模型的临床价值。结果筛选出4个影像组学特征,建立与胶质瘤分级显著相关的影像组学模型。该模型在训练组中AUC为0.723(95%CI:0.684~0.863),敏感度为75%,特异度为89%,校准度为0.120。在验证组中,AUC为0.800(95%CI:0.657~0.942),敏感度为73%,特异度为82%,校准度为0.561。决策曲线分析显示阈值概率在0.17%~0.99%时,影像组学模型对较低级别胶质瘤的分级能力优于将所有患者认为Ⅱ级及Ⅲ级。结论基于MRI T2WI序列图像建立的影像组学模型有助于区分较低级别胶质瘤的Ⅱ级及Ⅲ级,为患者制定手术方案和预后情况提供一种无创性技术手段。

Objective To evaluate the value of radiomics model in distinguishing grade Ⅱ and Ⅲ of gliomas from T2-weighted MRI images.Methods 159 gliomas patients(Mayo Clinic,October 2002-August 2011),who underwent non-enhanced MRI and tumor grades confirmation from the Cancer Genome Atla(TCIA)data portal,including grade Ⅱ(n=104)and Ⅲ(n=55)of gliomas.Patients were divided into training cohorts(n=111)and validation cohorts(n=48)in a ratio of 7:3.Gliomas were imported into the ITK-SNAP to manually delineate volume of interest(VOI)on T2-weighted images.The delineated data was imported into A.K software(Artificial Intelligence Kit v.3.1.0.R,GE Company)to extract tumor radiomics features.A total of 396 features were extracted,and the main features included 6 categories including Histogram,GLCM,GLSZM,RLM,Form Factor,Haralick.LASSO regression was used for feature screening.A formula was generated using a linear combination of selected features that were weighted by their respective LASSO coefficient.A radiomics score was calculated for each patient by the formula.The predictive accuracy of radiomics model was quantified by AUC in both cohorts.The calibration degree of the radiomics was evaluated by Hosmer-Lemeshow test.The clinical usefulness of the radiomics model was assessed by decision curve analysis.Results Four radiomics features were chosen to build a radiomics model that distinguished grade Ⅱ and Ⅲ of gliomas with an AUC,sensitivity,specificity and CD of 0.723,75%,89%,0.120 in training cohort;and 0.800,73%,82%,0.561 in the validation cohort,respectively.When the threshold probability of DCA is 0.17%-0.99%,the classification of lower grade glioma by radiomics model is better than that of all patients as grade Ⅱ and Ⅲ.Conclusion The radiomics model based on T2-weighted MRI images can distinguish grade Ⅱ and Ⅲ of lower grade gliomas,providing a non-invasive technique for developing a surgical plan and prognosis for gliomas patients.