基于多模态MRI影像组学模型预测MGMT甲基化阳性高级别胶质瘤1p/19q缺失状态  

Radiomics model based on multi-model MRI for predicting 1p/19q deletion status in high-grade gliomaswith positive-methylation MGMT

作　　者：赵伟 黄海燕 丁爽 罕迦尔别克·库锟 徐蕊 王云玲 ZHAO Wei;HUANG Hai-yan;DING Shuang(Imaging Center of the First Affiliated Hospital of Xinjiang Medical University,Urumchi 830054,China)

机构地区：[1]新疆医科大学第一附属医院影像中心,新疆乌鲁木齐830054

出　　处：《放射学实践》2024年第12期1545-1550,共6页Radiologic Practice

基　　金：中央引导地方科技发展专项资金项目(ZYYD2023D02);科技创新领军人才(2023TSYCLJ0027)。

摘　　要：目的:基于多模态MRI构建影像组学模型,无创性预测MGMT甲基化阳性高级别胶质瘤的1p/19q缺失状态。方法:回顾性分析2021年9月-2023年9月在本院经手术病理证实的106例高级别胶质瘤患者的完整临床和影像资料。所有肿瘤为MGMT甲基化阳性,其中合并1p19q共缺失者33例,非1p19q共缺失者73例。将所有患者按照7∶3的比例随机分为训练集和测试集。分别在T_(1)WI、T_(2)WI、T_(2)-FLAIR和CE-T_(1)WI四个序列的图像上,沿着肿瘤边缘逐层勾画ROI并生成容积感兴趣区(VOI)后提取影像组学特征。应用主成分分析(PCA)方法进行特征降维,并应用方差分析方法进行特征筛选,随后分别采用自编码器(AE)、逻辑回归(LR)、随机森林(RF)和支持向量机(SVM)四种机器学习算法构建预测MGMT甲基化阳性合并1p/19q共缺失状态的影像组学模型。采用受试者工作特征曲线评估各模型的诊断效能。结果:AE影像组学模型在预测高级别胶质瘤MGMT甲基化阳性合并1p/19q缺失状态表现出较高的AUC,其在训练集和测试集中的AUC分别为0.924和0.864;而LR、RF和SVM影像组学模型在训练集中的AUC分别为0.950、1.000和0.951,在测试集中的AUC分别为0.777、0.773和0.786。结论:基于多模态MRI影像组学模型可以有效预测MGMT甲基化阳性高级别胶质瘤的1p/19q缺失状态。Objective:The purpose of this study was to construct a radiomics model based on multi-model MRI for noninvasively predicting the 1p/19q deletion status in high-grade gliomas with methylation-positive MGMT.Methods:106 high-grade glioma patients who underwent surgery and were confirmed by pathological biopsy in the Department of Neurosurgery of the First Affiliated Hospital of Xinjiang Medical University from September 2021 to September 2023 were retrospectively analyzed,of which 33 patients with MGMT methylation-positive combined 1p19q co-deletion and 73 patients with MGMT methylation-positive combined 1p19q non-co-deficiency.All patients were randomly divided into a training set and a testing set at a ratio of 7∶3.T_(1)WI,T_(2)WI,T_(2)-FLAIR,and CE-T_(1)WI sequences were selected to outline the tumor region of interest(ROI)layer by layer along the tumor margin and generate the volume of interest(VOI)to extract the radiomics features.Principal component analysis(PCA)was applied for dimensionality reduction,and ANOVA method was used for further feature selecting.And then,four machine learning method including auto-encoder(AE),logistic regression(LR),random forest(RF),and support vector machine(SVM)were respectively used to build the radiomics models for predicting MGMT methylation-positive combined with 1p/19q co-deletion status.The diagnostic efficacy of each model was assessed by ROC curve analysis.Results:Among the four radiomics models,the AE radiomics model was the optimal model for predicting the 1p/19q deletion status in high-grade gliomas with positive-methylation MGMT,with AUC of 0.924 and 0.864 in the training set and test set,respectively;and the AUCs of the other three radiomics models(LR,RF,and SVM)were 0.950,1.000 and 0.951 in the training set,and 0.777,0.773 and 0.786 in the test set,respectively.Conclusion:Multi-model MRI radiomics-based model can effectively predict MGMT methylation positivity combined with 1p/19q deletion status in high-grade gliomas.

关 键 词：脑胶质瘤 影像组学 磁共振成像 O6-甲基鸟嘌呤-DNA甲基转移酶 1p/19q 

分 类 号：R445.2[医药卫生—影像医学与核医学] R739.41[医药卫生—诊断学]