基于模型的乳腺X线图像胸肌分割算法研究  

Study on model-based pectoral-muscle segment algorithmin mammograms

作　　者：徐伟栋[1] 王小英[2] 夏顺仁[1] 严勇[1] 

机构地区：[1]浙江大学生物医学工程教育部重点实验室,浙江杭州310027 [2]浙江大学校医院,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2005年第3期427-432,共6页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(60272029);浙江省自然科学基金资助项目(M603227).

摘　　要：针对目前的乳腺X线图像胸肌分割算法自适应能力较弱的问题,提出了一种基于胸肌模型的新算法.该算法将一组不同尺寸的感兴趣区(ROI)作用到乳腺X线图像,在每个 ROI中用迭代阈值法得到最优阈值组成一条曲线,并计算出与该最优阈值曲线对应的局部均方差曲线,根据提出的近似真实乳腺 X线图像胸肌模型的特征,自动确定乳腺X线图像中胸肌区域的最佳分割阈值.最后使用两段直线粗拟合和多边形精拟合,提取阈值化后的胸肌边界.实验结果表明该算法可以显著地提高胸肌分割的精度,并且具有较好的鲁棒性.To overcome the low adaptability of pectoral-muscle segmentation in mammograms, a novel model-based algorithm was presented. Regions of interest (ROI) with different sizes were applied on a mammogram respectively, and optimal thresholds were calculated with iterative thresholding algorithm in each of ROI. Composing an optimal threshold curve, thresholds generated a local mean-square deviation curve accordingly. Based on two pectoral muscle models, the optimal segment threshold of the pectoral muscle was extracted from the optimal threshold curve and local mean-square deviation curve. By using two-segment line fitting and polygon approaching technique, the edge of the pectoral muscle was obtained. Experimental results show that this algorithm has higher accuracy and adaptability than conventional methods.

关 键 词：感兴趣区(RO1) 乳腺X线图像 胸肌分割算法 直线拟合 多边形拟合 

分 类 号：R318.04[医药卫生—生物医学工程]