基于MRI-T_(1)增强影像组学特征分析预测脑膜瘤p53基因表型

Predicting p53 gene phenotype in meningioma based on radiomics analysis of enhanced MRI T_(1)-weighted imaging

目的:探究基于MRI-T_(1)增强图像影像组学特征分析在预测脑膜瘤p53基因表型中可行性及存在的价值。方法:搜集本院2016年6月-2020年6月经术后病理及免疫组化证实脑膜瘤患者80例(p53基因野生型与突变型各40例)。将所有患者T_(1)增强图像导入基于MATLAB R2014a平台开发IBEX图像处理软件,由两位影像科硕士研究生在上级医师指导下勾画肿瘤感兴趣区(ROI)并提取影像组学特征。对特征数据进行降维处理,再分别采用逻辑回归、决策树、支持向量机以及自适应增强4种分类学习器构建脑膜瘤p53基因表型预测模型,同时使用受试者工作特征曲线(ROC)评价各种模型预测效能。选取上述4种分类学习器中表现最佳的一种,依次构建影像特征模型、联合模型并建立脑膜瘤p53基因突变风险量化评估模型(Nomogram模型)。结果:组间存在5个具有非零系数影像组学特征,4种分类学习器中SVM模型预测效能最佳,其训练集与验证集AUC(area under curve)分别为0.894和0.729;进一步纳入肿瘤强化均匀度、扩散是否受限这两个影像特征后得到联合模型AUC为0.954。Nomogram模型显示脑膜瘤出现扩散受限、强化不均匀表现且Radiomics值越高时患者为p53基因突变型可能性越大。结论:基于MRI-T_(1)增强图像影像组学特征分析结合影像特征建立联合模型对脑膜瘤p53基因表型具有较好预测价值,而最终构建Nomogram预测模型能对脑膜瘤p53基因突变风险进行量化评分的同时也为个体化评估脑膜瘤内生物学特性提供了有用参考依据。

Objective:To explore the feasibility and application value of radiomics analysis in predicting p53 gene phenotype in meningioma based on enhanced T 1-weighted imaging.Methods:A total of 80 patients with meningioma(40 wild-type and 40 mutant-type patients),which were confirmed by pathology and immunohistochemistry,were recruited from June 2016 to June 2020 in the First Affiliated Hospital of Kunming Medical University.The postcontrast T 1WI of all patients were analyzed using IBEX imaging analysis software,which was developed on MATLAB R2014a platform.Then,the regions of interest(ROI)of the tumor were delineated by two graduate students under the guidance of an experienced radiologist,and the radiomics features were extracted.After reducing the dimensions of feature data,four classification learning machines,including logistic regression,decision tree,support vector machine(SVM)and adaptive enhancement,were respectively used to construct prediction model of the p53 gene phenotype.Meanwhile,the receiver operating characteristic curves(ROC)were used to evaluate the predictive value of four models.Finally,the best classification learning machine was used to construct the image feature model,the combined model and the Nomogram for predicting meningioma with p53 gene phenotype.Results:Five radiomics features showed non-zero coefficients among groups.The SVM had the best predictive value,which the AUC(area under curve)of training set and validation set were 0.894 and 0.729,respectively.Furthermore,the AUC of combined model increased to 0.954 while added the two more image characteristic,which the uniform enhancement and diffusion of meningioma were limited.In addition,the Nomogram showed that the greater possibility of p53 gene mutation,when meningioma patients appeared limited diffusion,heterogeneous enhancement and higher Radiomics values.Conclusion:The combined model had a good performance for predicting p53 gene phenotype in meningioma,and the Nomogram can quantify the risk of meningioma p53 gene mutation,and also provide a useful reference for individual assessment of the intrinsic biological characteristics in meningioma.