基于双视角乳腺X线图像的微钙化簇检测

Detection of microcalcification clusters based on two mammographic views

本文提出了一种通过将两个视角的图像信息进行综合分析的方法,以降低乳腺X线图像中微钙化簇检测中的假阳性率。基于微钙化簇通常出现在两个视角图像中这一事实,文中提出了一种微钙化簇匹配技术:首先把MLO视角探测到的可疑微钙化簇通过空间位置天系找到CC视角中与其相对应的病变区域,形成微钙化簇对;然后对匹配后的每对微钙化簇提取面积、形态、灰度等簇特征,通过特征之间的相似度判断所对应微钙化簇的真伪。实验结果表明:本文的微钙化簇检测算法较单视角检测特异度增加了15%。

To reduce the number of false positive rates, we have developed a new method for computer aided detection of microcalcification clusters using joint analysis of two views of the same breast. A cluster matching technique is proposed based on that the clustered microcalcifications （MCCs for short） should emerge in both views. The algorithm first links a suspicious cluster located in the MLO view with the corresponding location in the CC view using their spatial information and forms a paired cluster, then each cluster candidate is characterized by its single-view features such as size, shape and intensity. Finally a similarity function is built between the features to estimate whether they are true clusters or not. Experiment results show that the specificity measure of the proposed system has increased by 15% compared with the detection algorithm based on single view.