乳腺肿瘤超声图像的反应扩散水平集分割  

Segmentation of Breast Tumor Ultrasound Images Using Reaction Diffusion Level Set

作　　者：陈帅[1] 张麒[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200072

出　　处：《自动化仪表》2015年第9期34-37,41,共5页Process Automation Instrumentation

基　　金：国家自然科学基金青年资助项目(编号:61401267);上海市自然科学基金资助项目(编号:12ZR1444100);上海市教委人才计划"晨光计划"资助项目(编号:11CG45);上海市教委科研创新项目(编号:12YZ026)

摘　　要：针对乳腺肿瘤超声图像分割,提出一种改进的反应扩散(RD)水平集分割算法。先使用Gabor各向异性扩散模型进行滤波,由此构造边界停止函数;再将该函数融入RD水平集演化方程,以控制曲线的演化得到乳腺肿瘤的边界。采用该方法和传统RD方法对77例病人的111幅乳腺超声图像进行分割实验,分割准确率分别为98.5%和98.0%,真阳性率分别为88.2%和82.7%,与金标准之间的均方根误差分别为3.6和4.6像素。结果表明,该改进算法可获得更加准确的乳腺肿瘤分割结果。For segmentation of breast tumor ultrasonic images, the improved reaction diffusion （ RD ） level set segmentation algorithm is proposed. Firstly, the Gabor anisotropic diffusion model is used for filtering, to construct edge stopping function; then the function is integrated into RD level set evolution equation, for controlling the evolution of curves to obtain the boundary of breast tumor. Segmentations of 111 breast ultrasonic images from 77 patients are conducted with the proposed method and the traditional method, the segmentation accuracies are 98.5% and 98.0% respectively ; the true positive rate are 88.2% and 82.7% ; and the mean square errors are 3.6 and 4.6 pixels. The results show that the proposed improved algorithm provides more accurate segmentation for breast tumor.

关 键 词：边界停止函数 反应扩散 水平集演化 图像分割 乳腺超声图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]