高分辨率颅脑CT图像纹理统计图谱的创建与应用  

Creation and Application of Statistical Texture Atlas Based on Brain High-Resolution Images

作　　者：刘伟[1] 韩真[1] 李传富[1,2] 冯焕清[1] 

机构地区：[1]中国科学技术大学电子科学与技术系,合肥230027 [2]安徽中医学院第一附属医院影像中心,合肥230031

出　　处：《数据采集与处理》2009年第4期524-530,共7页Journal of Data Acquisition and Processing

基　　金：国家自然科学基金(60771007)资助项目

摘　　要：针对脑出血和脑肿瘤的自动检出应用,提出了一种创建高分辨率颅脑CT图像纹理统计图谱的方法。采用图像局部直方图的多阶矩特征结合多分辨率策略提取颅脑CT图像的纹理特征,并在特征中融合边缘与区域信息。在创建统计图谱时,对经过预处理的样本图像使用D em ons方法进行非刚性配准,并提取多分辨率纹理特征及其统计参量。检测病变时将待测样本的纹理特征向量与图谱比较,并以M aha lanob is距离作为病变发生概率的度量进行阈值分割。实验表明,本文方法对均匀密度和混杂密度型颅脑病变均有较好的诊断效果,且计算复杂度较低。Aimed at most cases of cerebral tumor usually reveal diagnostic information a statistical and hemorrhage, texture patterns of tissue texture atlas method is presented hased on high-resolution CT images and used for brain lesion detection. Firstly, the texture of each single normal brain is represented as feature vectors of geometric moment invariants. These normal sample images are well-registered both rigidly and non-rigidly. Then, the distribution parameters of these feature vectors is established to generate a statistical texture atlas. For the lesion detection, Mahalanobis distance between the texture vector space of abnormal brain images and the normal brain atlas provides an evidence of abnormity. Experimental results indicate that the statistical texture atlas can distinguish details of brain tissues and has strong detection power of uniform and mixed density lesions.

关 键 词：病变检出 统计图谱 纹理特征 MAHALANOBIS距离 

分 类 号：R318.04[医药卫生—生物医学工程]