基于改进C-V模型的肾脏CT图像分割方法  

Segmentation of kidney CT images based on an improved C-V model

作　　者：张品[1] 梁艳梅[1] 常胜江[1] 

机构地区：[1]南开大学现代光学研究所,光学信息技术科学教育部重点实验室,天津300071

出　　处：《光电子．激光》2013年第3期602-607,共6页Journal of Optoelectronics·Laser

基　　金：国家自然科学基金(60677012);教育部博士点基金(20090031110033);天津市应用基础与前沿技术研究计划(09JCZDJC18300)资助项目

摘　　要：生物组织的自动分割是计算机辅助诊断和病变检测的关键步骤。在腹腔CT图像中,肾脏组织本身的灰度不均匀性使得传统C-T模型无法准确实现肾脏的分割。为了解决上述问题,本文结合图像全局和局部统计信息改进了传统的C-V模型。基于先验知识,提出了描述肾脏组织皮质特征的数学表达式。选择感兴趣区域,在预处理阶段获得了CT图像中肾脏的大致初始轮廓。随后,应用C-V模型进行轮廓演化时引入局域信息,提高了C-V模型的局部适应性。实验结果表明,与现有方法相比,本文的方法的结果更接近于人工分割结果,其肾脏分割结果的Dice系数平均值为94.0%。Automatic segmentation of biological tissues is a vital step in computer-aided diagnosis and pa thology detection. In abdominal computed tomography （CT） images, renal tissues have intensity inhomo- geneity and cannot be segmented accurately by the traditional C-V model. In this paper,according to the characteristics of renal tissues in CT images, an improved C-V model combining global and local information is proposed to solve the problem,which is able to get better segmentation results. At first,based on the prior knowledge of renal tissues in CT images, the regions of interest including renal tissues are obtained to speed up the operation. At the same time,in the pre-processing stage,cortical properties are described in mathematical expressions and the renal initial contour is also obtained to locate the renal tissues roughly. Then, in order to improve the local adaptability for extracting renal tissues, the local information is introduce in the C-V model within the contour evolution in the regions of interest,which pro vides a more reliable basis for contour convergence on renal boundaries. Compared with the available methods, the experimental results show that the kidney segmentation of our proposed method is closer to the ground truth and accordingly the dice coefficient of our kidney segmentation is about 94. 0%.

关 键 词：计算层析(CT) 肾脏分割 C—V模型 皮质特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]