基于DCE图像优化胶质瘤自动分级纹理模型  

The influence of texture retrieving method for support vector machine with DCE MR images to grade glioma

作　　者：南海燕 杨洋 颜林枫 张欣 王文 崔光彬 NAN Hai-yan;YANG Yang;YAN Lin-feng;ZHANG Xin;WANG Wen;CUI Guang-bin(Department of Radiology,Tangdu Hospital,the Fourth Military Medical University,Xi’an 710038,China)

机构地区：[1]空军军医大学唐都医院放射诊断科,西安710038

出　　处：《磁共振成像》2018年第10期737-741,共5页Chinese Journal of Magnetic Resonance Imaging

基　　金：陕西省社会发展科技攻关项目(编号:2014JZ2-007);空军军医大学(第四军医大学)唐都医院科技创新发展基金(编号:2016LCYJ001)~~

摘　　要：目的探讨不同纹理模型和灰阶对基于动态对比增强磁共振图像(dynamic contrast enhancement magnetic resonance imaging,DCE-MRI)的支持向量机的胶质瘤自动分级影响。材料与方法收集经磁共振扫描且经病理证实为胶质瘤Ⅱ、Ⅲ、Ⅳ级的患者共117例,计算DCE-MRI图像血流动力学参数(NordicICE 4.0),利用不同纹理模型和灰阶提取参数图肿瘤区域相应纹理特征。支持向量机递归特征消除算法选择特征后,输入线性SVM对胶质瘤级别进行分类并使用留一法交叉验证。分类结果使用Graphpad Prism 6统计软件分析。结果灰阶对分类效能的影响差异无统计学意义(P=0.1589),纹理模型对分类效能的影响差异存在统计学意义(P<0.0001)。在使用灰度共生矩阵(gray-level cooccurrence matrix,GLCM)提取纹理特征并且灰阶为32和256时,分别选取前22个和前17个特征所得分类正确率最高(正确率=0.79)。结论基于DCE图像纹理对支持向量机胶质瘤分级中,纹理模型GLCM结合特征选择是胶质瘤分级的最优方案,并推荐在后期研究中使用。Objective:This study aimed to investigate the influence of texture retrieving model of dynamic contrast enhancement magnetic resonance imaging(DCEMRI)for SVM-based glioma grading.Materials and Methods:One hundred and seventeen glioma patients(pathology confirmed grade II/III/IV)receiving MRI scans were retrospectively included.The tumor image texture attributes were retrieved using three common texture retrieving models,including GLCM,GLRLM,and GLSZM models.Model-derived features were input into linear SVM scheme and SVMrecursive feature elimination(SVM-RFE)feature selecting strategies to compare the model-dependent grading accuracies that were further tested with leave-one-out crossvalidation(LOOCV).Classification results were further analyzed by Graphpad Prism 6.Results:Gray level had no significant influence on classification performance(P=0.1589).Texture model had obviously contracted(P=0.0001).GLCM performed best in combination with gray level 32 and 256 by using the top 22 and 17 attributes,respectively(accuracy=0.79).Conclusions:When using DCE-MRI image textures based SVM classification of gliomas,GLCM model in combination with feature selection performed best and should be recommended for preoperative glioma grading.

关 键 词：纹理模型 灰阶 支持向量机 动态对比增强 磁共振成像 神经胶质瘤 

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]