【摘要】 为了有效辅助脑胶质瘤的精确分级，提出一种基于影像组学的脑胶质瘤分级方法。首先，提取脑胶质瘤患者的T1加权磁共振图像的影像组学特征，在十折交叉验证框架下采用Boruta算法进行特征筛选得到重要特征；其次，采用随机森林将筛选得到的重要特征构建脑胶质瘤预测分级模型；最后，对脑胶质瘤预测分级结果进行评价与分析，用统计学方法检验不同级别特征的差异性。实验结果表明，提出方法的平均准确率达到84.75%,平均特异性达到85.32%,平均敏感性达到84.19%,平均受试者操作特征曲线下面积(Area Under Curve, AUC)达到0.92。更多还原

【Abstract】 In order to effectively assist the accurate grading of brain glioma, a grading method of brain glioma was proposed based on radiomics. Firstly, ten-fold cross-validation framework is used to extract the radiomics features from T1-weighted magnetic resonance images of patients with glioma, and Boruta algorithm is used for feature selection. Secondly, random forest is used to construct the predictive model of glioma grading with the selected important features. Finally, the grading results of brain glioma are evaluated and analyzed, and the differences of features at different grades are tested with statistical methods. Experimental results show that the average accuracy of the proposed method is 84.75%, the average specificity is 85.32%, the average sensitivity is 84.19%, and the average area under the operating characteristic curve(AUC) of the model is 0.92.更多还原