乳腺X线图像肿块建模与分割

Mass Modeling and Segmentation in Mammogram

提出了一种基于模型分析与均值漂移聚类的乳腺肿块分割方法.该方法根据肿块的临床特征表现建立了肿块的数学模型,并通过多重滤波实现肿块的准确定位.在此基础上,结合均值漂移算法获得的像素点集合,筛选出初始肿块.最后利用无边缘活动轮廓模型准确分割出肿块.实验采用通用的MIAS数据库进行算法性能测试,结果表明本文方法在保证较低假阳性率的同时,肿块检测真阳性率高于形态学成分分析方法.此外,本文方法分割出的肿块边界完整,可满足临床检验与诊断需求.

A novel method of mass segmentation in mammogram is proposed in this paper. First, a mathematical model is presented based on the clinical features of the mass and a multi-filtering method is used to detect the mass＇ location. Then, according to the result of clustered pixels, which is got by mean shift algorithm, rough handling masses could be obtained. Finally, the active contour without edge model is applied to refine the rough-wrought masses. Experiments implemented on the public MIAS database indicate that the proposed method can achieve better true positive rate than the morphological component analysis method with low false positive rate. In addition, the proposed segmentation method could emerge complete boundary, which could meet the clinical examination and diagnosis demand.