磁共振图像的纹理分析在儿童常见后颅窝肿瘤鉴别诊断中的应用

Utility of texture analysis of magnetic resonance imaging in differential diagnosis of common pediatric cerebellar tumors in children

目的通过对3种儿童常见后颅窝肿瘤的磁共振图像进行纹理分析，并用支持向量机（SVM）评价纹理特征。方法回顾性分析2008年5月至2015年8月郑州大学第一附属医院经手术病理证实的76例儿童后颅窝肿瘤，对提取的三种肿瘤的感兴趣区进行纹理分析。结果利用灰度共生矩阵提取出的5个纹理参数中至少有3个参数在上述两种不同类别的数据集之间具有统计学意义（P〈0．05），对于所有不同类别的数据集，ROI区域灰度值的方差参数均有统计学意义。室管膜瘤与髓母细胞瘤、室管膜瘤与星形细胞瘤、髓母细胞瘤与星形细胞瘤具有统计学差异特征的SVM测试准确率分别为86．15％±4．16％、73．63％±5．82％、74．32％±5．85％。结论纹理特征的分析可以提供更多量化信息特征，为儿童后颅窝肿瘤影像学的鉴别诊断提供了新的思路和方法。

Objective To analyze magnetic resonance images of the posterior cranial fossa tumors, and evaluate the texture characteristics of three kinds of children＇s posterior cranial fossa tumors by support vector machine （SVM）. Methods From May 2008 to August 2015, a total of 76 patients with children＇s posterior cranial fossa tumors in First Affiliated Hospital of Zhengzhou University were retrospectively analyzed, texture analysis was performed on the three types of tumors. Results The 5 texture parameters extracted from the gray level co-occurrence matrix had at least 3 parameters which had statistically significant difference between the two different classes of data sets （P 〈0.05）. For all kinds of data sets, the variance parameters of gray value of ROI region had statistically significant differences. The accuracy rate of SVM test in the remarkable difference of the ependymomas and medulloblastomas, the ependymomas and astrocytomas and the medulloblastomas and astrocytomas was 86.15% ± 4.16% , 73.63% ±5.82% and 74.32% ± 5.85%, respectively. Condusion The analysis of texture features can provide more quantitative information which could provide a new thought and method for the differential diagnosis of tumors in the posterior cranial fossa.