基于灰度共生矩阵的大鼠肝细胞癌、肝硬化结节超顺磁性氧化铁MR增强图像纹理特征分析  

Texture analysis of SPIO-enhanced MR imaging in rat models of hepatocellular carcinoma and hepatocirrhosis based on gray level co-occurrence matrix

作　　者：郭冬梅[1,2] 邱天爽[1] 康伟[1] 张丽[3] 

机构地区：[1]大连理工大学电子与信息工程学院,辽宁大连116024 [2]大连医科大学附属第二医院放射科,辽宁大连116027 [3]大连医科大学微生物教研室,辽宁大连116027

出　　处：《中国医学影像技术》2010年第3期563-566,共4页Chinese Journal of Medical Imaging Technology

基　　金：国家自然科学基金(30570475)

摘　　要：目的基于灰度共生矩阵分析大鼠肝细胞癌(HCC)结节及肝硬化超顺磁性氧化铁(SPIO)增强图像纹理特征。方法建立大鼠HCC及肝硬化模型,获取大鼠HCC、肝硬化结节SPIO增强图像,手动选取161个(HCC结节81个,肝硬化80个)感兴趣区(ROI),通过生成灰度共生矩阵,提取两组图像的二阶矩、对比度、相关、逆差矩、熵、方差6种纹理特征参数,并进行统计学分析。结果大鼠HCC结节呈相对均匀的高信号,较大HCC灶(>2 cm)呈高、低混杂信号,肝硬化结节在SPIO增强图像呈低信号。HCC组的相关及熵均值高于肝硬化组(P<0.05),二阶矩、对比度、逆差矩及方差的均值低于肝硬化组(P<0.05)。结论通过灰度共生矩阵提取的二阶矩、对比度、相关、逆差矩、熵、方差6种纹理特征参数,可进一步用于基于纹理特征的大鼠HCC、肝硬化结节SPIO增强图像的计算机辅助诊断。Objective To analyze the texture features of SPIO-enhanced MR imaging in rat models of hepatocellular carcinoma（HCC） and hepatocirrhosis with gray level co-occurrence matrix（GLCM）.Methods HCC and hepatocirrhosis models were established in rats.SPIO-enhanced MR images were obtained.A total of 161 regions of interests（ROIs,81 of HCC and 80 of hepatocirrhosis） were selected manually.Feature values as angular second moment,contrast,correlation,inverse difference moment,entropy,variance were extracted based on GLCM.The differences of feature values between two groups were statistically analyzed.Results In SPIO-enhanced MR images,hypointense signal changes were found in hepatocirrhosis,as well as hyperintensity in HCC nodules and intermixed intensity in larger HCC nodules.Correlation and entropy values of HCC group were higher than that of hepatocirrhosis group,while the angular second moment,contrast,inverse difference moment,and variance values were lower than hepatocirrhosis group.Conclusion The feature values based on GLCM could be used for the further computer aided diagnosis of SPIO-enhanced MR images in rat models of HCC and hepatocirrhosis.

关 键 词：纹理分析 灰度共生矩阵 肝硬化 癌 肝细胞 磁共振成像 

分 类 号：R445.2[医药卫生—影像医学与核医学] R332[医药卫生—诊断学]