摘要
目的 探讨基于多模态影像组学方法在脑胶质瘤高低分级中的应用价值.方法 选择258例脑胶质瘤患者作为观察对象,高分化脑胶质瘤210例,低分化脑胶质瘤48例.所有患者按9‥1的比例分为训练组和测试组,选择增强肿瘤区域作为感兴趣区域(ROI),通过Pyradiomics开源库提取T1、T2、T1ce和Flair 4 个模态MRI共428组影像组学特征.选出具有最高预测价值的影像组学特征,构建对数几率回归(LR)、支持向量机(SVM)和多层感知机(MLP)3种机器学习模型进行脑胶质瘤分级,并对测试组进行验证.结果 利用LR、SVM和MLP 3种机器学习算法构建的影像组学模型在训练集的曲线下面积(AUC)均>0.95,测试组均>0.90.基于LR构建的影像组学模型最优,其在测试组中的准确率、AUC、敏感度和特异度分别为92.0%、0.976、90.5%和100.0%.结论 基于多模态MRI影像组学特征结合机器学习分类模型可准确预测脑胶质瘤的高低分级.
Abstract
Objective To explore the application value of multimodal radiomics methods in the high and low grades of glioma to assist clinical diagnosis.Methods A total of 258 cases of glioma were analyzed,included 210 cases of well-differentiated glioma and 48 cases of poorly-differentiated glioma.Patients were randomly divided into training and testing groups in a ratio of 9‥1.The enhanced tumor Regions were selected as Regions of interest(ROI),and a total of 428 groups of radiomics features from four MRI modalities(T1,T2,T1ce and Flair)were extracted through the Pyradiomics open source library.Spearman correlation test and Least Absolute Shrinkage and Selection Operator(LASSO)were used to select the radiomics features with the highest predictive value in the training group.Then three Machine learning models including logistic regression(LR),Suppot Vector Machine(SVM)and Multilayer Perception(MLP)were constructed to grade glioma.The results were validated in the test group.Results Among the radiomics models constructed by LR,SVM,and MLP machine learning algorithms,the AUCs of all models were greater than 0.95 in the training set and greater than 0.90 in the testing machine.The radiomics model based on LR was the best,with accuracy,AUC,sensitivity and specificity of 92.0%,0.976,90.5%and 100.0%in the test set,respectively.Conclusion Machine learning classification model based on multi-modality MRI-based radiomics features can accurately predict the grade of brain gliomas.
维普
万方数据