肺结节CT图像纹理特征的初步研究  

Application of Gray-scale Texture Feature in the Diagnosis of Pulmonary Nodules on CT Imaging

作　　者：向子云[1] 史长征[2] 周洁[1] 詹勇[1] 罗良平[2] 

机构地区：[1]深圳市龙岗区人民医院,广东深圳518172 [2]暨南大学附属第一医院医学影像中心,广州510630

出　　处：《CT理论与应用研究（中英文）》2013年第1期155-160,共6页Computerized Tomography Theory and Applications

摘　　要：目的:探讨图像纹理特征分析法在肺结节诊断中的作用。方法:收集104例经临床手术病理证实的肺结节患者,采用水平集模型自动分割、提取肺结节,用灰度共生矩阵法对肺结节的能量、相关、对比度、逆差距和熵等图像纹理特征进行提取,将良、恶性结节的纹理特征进行比较。结果:肺良恶性结节的能量、相关、对比度、逆差距和熵均存在统计学差异(P<0.05),其中肺良性结节的能量、逆差距、相关性均大于恶性结节,良性结节对比度及熵的均值小于恶性结节。结论:肺良性结节纹理灰度分布均匀、纹理规则、局部灰度反差不大,而肺恶性结节纹理灰度分布不均匀、纹理杂乱、灰度反差明显,因此图像纹理特征对肺结节的诊断有重要的价值。Objective： To explore the value of image texture analysis in the diagnosis of pulmonary nodules. Methods： 104 cases of pulmonary nodules were collected, all of which were confirmed by pathology. The level set model was used to extract and segment the lung nodule automatically. The features of energy, contrast, correlation, homogeneity and entropy in pulmonary nodules were calculated by gray-level co-occurrence matrix, which were compared between the benign and malignant lung nodules. Results： The characteristics of energy, contrast, correlation, homogeneity and entropy have significant statistical difference between benign and malignant pulmonary nodules （P 〈 0.05）. The means of energy, homogeneity, correlation in benign nodules are greater than that in malignant nodules while the means of contrast and entropy in benign nodules are less than that of malignant nodules. Conclusion： The texture characteristics in pulmonary benign nodules are gray relatively uniform, local texture regular, relatively little contrast in gray, while that in pulmonary malignant nodules are local texture tangle some, gray uneven distribution, marked gray contrast, so the image texture features have important value in the diagnosis of lung nodules.

关 键 词：肺结节 纹理特征 体层摄影术 X线计算机 灰度共生矩阵 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]