Image segmentation is significant in the quantitative and qualitative analysis of medical ultrasound images,and directly affects the subsequent analysis and processing works.The ultrasound image of the breast tissue which has the characteristics of complex image texture,obvious noise,and low contrast due to the specificity of the breast tissue is difficult to be segmented automatically and accurately,therefore the diagnosis of breast tissue depends more on the observation of doctors.Herein,we propose an ultrasound image segmentation method based on superpixel and observer biased fuzzy clustering technology.The compactness adaptive simple linear iterative clustering is introduced to generate superpixels;the computation complexity is reduced through the initial segmentation;the feature vector set is calculated with the data from superpixels;the cluster validity analysis and observer biased fuzzy C-means with focal point are introduced to cluster the feature vector set,finally realizing the effective segmentation of target region.Using this method,we perform the image segmentation experiment on 20 frames of breast ultrasound images,and quantitatively evaluate the segmentation results according to the region-based evaluation criteria.The evaluation result shows that we can obtain better segmentation results using the proposed method which achieves (92.87±2.98)％ of true positivity,(11.05±2.75)％ of false positivity,and (83.39±3.64)％ of similarity.This research is of great significance in the study of the assistance method for ultrasound diagnosis of breast mass.%图像分割在医学超声图像的定量、定性分析中均扮演着十分重要的作用,并直接影响到后续的分析、处理工作.乳腺组织的特殊性导致了其超声图像纹理复杂、噪声明显、对比度较低,临床应用难以准确自动分割,诊断较依赖于人工观测.针对此问题提出一种基于超像素和模糊聚类技术相结合的图像分割算法.采用紧密度自适应的简单线性迭代聚类产生超像素,实现初步划分,减小计算量;计算各超像素的特征向量组成集合;采用聚类有效性分析和有偏观测模糊C均值聚类将特征向量分类,实现目标区域的有效分割.采用此方法对20帧乳腺超声图像进行图像分割实验,采用基于区域的评价准则对分割结果进行量化评估,评估结果表明真阳性为92.87％±2.98％,假阳性为11.05％±2.75％,相似性为83.39％±3.64％,取得了较好的分割结果.对乳腺肿块超声诊断的辅助方法研究具有探索意义.
展开▼
