摘要
目的研究MRI纹理分析技术在大鼠胶质瘤模型肿瘤分级中的应用,探讨不同纹理特征对胶质瘤分级诊断效能的差异。材料与方法选取C6胶质瘤细胞荷瘤大鼠32只,于接种后1～4周获取肿瘤MRI图像及病理级别,使用MaZda软件计算病灶纹理特征,利用费希尔系数、特征子集选择、互信息等方法选择特征,通过主成分分析、线性/非线性判别分析联合K近邻算法、人工神经网络特征分类,筛选出最佳分析方法、分类序列及纹理特征组。将最佳特征组纳入Logistic回归模型进一步分析,采用受试者工作特性曲线评价不同纹理特征的诊断效能。结果利用分类错误概率联合平均相关系数、非线性判别分析联合人工神经网络可获得较低的误判率。在最佳分类序列T2WI上提取的游程长非均匀度分级诊断效能最佳,受试者工作特性曲线下面积为0.839±0.049(P<0.01),截断值为245.44,对应的敏感度为0.824,特异度为0.733。结论 MR T2WI提取的纹理特征对大鼠高、低级别胶质瘤有较好的鉴别能力,其中以游程长非均匀度的分级诊断效能最佳。
Purpose To study the application of MRI texture analysis technique in tumor grading of rat glioma model, and to explore efficacy difference of different texture features in grading diagnosis of glioma. Materials and Methods Thirty-two C6 glioma cell tumor-bearing rats were selected, and tumor MRI images and pathological grades were obtained 1 to 4 weeks after inoculation. MaZda software was used to calculate lesion texture features, and Fisher coefficient, feature subset selection, mutual information and other methods were adopted to select features. The optimal analysis method, classification sequence and texture feature groups were screened out drawing on principal component analysis, linear/nonlinear discriminant analysis combined with K-nearest neighbor algorithm and artificial neural network feature classification. The best feature group was included into the logistic regression model for further analysis. Moreover, the receiver operating characteristic curve was adopted to evaluate the diagnostic efficiency of different texture features. Results A lower false positive rate can be realized using classification error probability combined with average correlation coefficient, and nonlinear discriminant analysis combined with artificial neural network. Diagnostic efficiency of run-length non-uniformity grading extracted on the optimal classification sequence T2WI was the best, with area under the receiver operating characteristic curve being 0.839±0.049 （P〈0.01）, cut-off value was 245.44, corresponding sensitivity 0.824, and specificity 0.733. Conclusion The texture features extracted by magnetic resonance T2WI have a good discriminant ability for high- and low-grade gliomas in rats, among which the grading diagnosis with run-length non-uniformity demonstrates best efficiency.
作者
张欢欢
林丽萍
王娇燕
ZHANG Huanhuan;LIN Liping;WANG Jiaoyan(Department of Radiology,the Fifth People＇s Hospital of Shanghai,Fudan University,Shanghai 200240,China)
出处
《中国医学影像学杂志》
CSCD
北大核心
2018年第11期808-814,共7页
Chinese Journal of Medical Imaging
关键词
神经胶质瘤
磁共振成像
病理学
外科
图像处理
计算机辅助
疾病模型
动物
大鼠
Wistar glioma
Magnetic resonance imaging
Pathology
surgical
Image processing
computer-assisted
Disease models
animal
Rats
Wistar
