基于相位和GGVF的水平集乳腺超声图像分割  

Phase and GGVF-based level set segmentation of breast ultrasound images

作　　者：高梁[1] 刘晓云[1] 陈武凡[1,2] 

机构地区：[1]电子科技大学自动化工程学院,成都611731 [2]南方医科大学生物医学工程学院,广州510515

出　　处：《仪器仪表学报》2012年第4期870-877,共8页Chinese Journal of Scientific Instrument

基　　金：国家973重点基础研究发展规划项目(2010CB732501)资助

摘　　要：提出一种基于相位和广义梯度矢量流(generalized gradient vector flow,GGVF)的水平集分割方法,并用于乳腺超声图像的肿瘤分割。首先,在频域空间,用Cauchy核替代Log-Gabor作为正交滤波器对图像进行滤波,提取来自于单演信号的多尺度图像特征,引入相位一致的思想将多尺度特征结合起来进行边界检测;然后,在此基础上,利用相位一致梯度图定义了一个基于相位的速度停止项函数,同时改进了GGVF;最后,将得到的速度停止项和梯度矢量流融入到水平集演化方程中来控制曲线的演化,获得乳腺肿瘤的边界。实验结果表明,使用该分割方法可获得比现有方法更好的乳腺肿瘤分割结果。This paper presents a novel algorithm based on phase and generalized gradient vector flow（GGVF） for the segmentation of breast tumors in ultrasound images.First,in frequency domain,we use Cauchy kernels,as a better alternative for the commonly used log-Gabor,as a pair of quadrature filters to filter the images and extract the multi-scale image features from the monogenic signal.The idea of phase congruency is then introduced to integrate the multi-scale features for the edge detection.Second,on this basis,phase congruency gradient map is used to define a new speed stop term function and improve the GGVF model.Finally,the defined speed stop term and GGVF are combined into the level set evolution equation to guide the curve propagation and obtain the tumor boundaries.Experiments performed on breast ultrasound images show that the proposed method has better tumor segmentation performance than existing methods.

关 键 词：局部相位 广义梯度矢量流 水平集 分割 乳腺超声图像 

分 类 号：R318.08[医药卫生—生物医学工程]