摘要
提出一种基于邻域信息和高斯加权卡方距离的脊椎MR图像椎体的自动分割方法。由于成像过程中存在噪声和各向异性的影响,单个像素的灰度值对噪声敏感,为此采用5像素×5像素窗口,提取每个像素点邻域内的空间-灰度特征,该特征对噪声具有较强的鲁棒性。利用高斯加权的卡方距离度量两个像素的相似性,构造一种全新的相似度矩阵;而单一的尺度参数存在一定局限性,所以引入一种自适应的局部收缩因子,完成脊椎MR图像椎体的自动分割。实验结果表明,新算法克服了传统方法中常见的过分割和欠分割现象,覆盖率均在96%以上;分割的正常和退行性改变椎体光滑且清晰,具有分割结果准确、鲁棒性强的优点。作为一种一般性的分割方法,该算法可以拓展到其他器官的分割中。
In this article,we propose a novel approach for solving the segmentation based on local neighborhood information and Gaussian weighted Chi-square distance of the vertebral bodies from 2D sagittal magnetic resonance images of the spine automatically.Due to the noise and anisotropic factors＇ existence,The pixel intensity in the same tissue is sensitive to noise and varies sharply.Thus,a box of 5×5 centered on each pixel is used to calculate local spatial-intensity feature which is robust to the noise.Then,the Gaussian function and Chi-square test are used to evaluate the distance between the two pixels for the sake of creating a novel affinity matrix with the vital spatial structure and intensity feature of the image.Rather than selecting a single scaling parameter,an adaptive local scaling parameter is used to refine the graph-based segmentation.Avoiding the phenomenon of over-segmentation and under-segmentation,the encouraging results and the rate of the coverage more than 96% indicate that our new algorithm has high accuracy and strong robustness and can segment the vertebral bodies smoothly and clearly if the patient has vertebral body lesions.It is a general method for segmenting objects and can be developed to segment other tissues and organs.
出处
《中国生物医学工程学报》
CAS
CSCD
北大核心
2011年第3期357-362,共6页
Chinese Journal of Biomedical Engineering
基金
国家重点基础研究发展(973)计划(20100CB732500)
国家自然科学基金重点项目(30730036)
国家自然科学基金(31000450)
关键词
高斯核函数
卡方距离
局部收缩
相似度矩阵
Gaussian function
Chi-square distance
local scaling
affinity matrix
