Using Mean-shift Clustering and Statistical Analysis for Breast Mass Segmentation and Category

Abstract

[Research objective] Breast mass segmentation is a crucial step in computer-aided diagnosis systems. The objective of this study is to propose a breast mass segmentation method using mean-shift clustering and statistical analysis. [Research method] The proposed method is described as follows. Firstly, the mean-shift algorithm splits the breast mass image into different clusters. Subsequently, the clusters of interest are identified by finding the overlapping of the split clusters and a circle covering the mass. Finally, Levene’s test and Welch’s test are both used to determine whether or not to merge clusters among clusters of interest. [Research result] The experimental results show that the proposed can perform mass segmentation effectively even if the margin of a mass is obscure and difficult to identify. In addition, the experiments also show that the Zernike moments method and the k-means method are effective methods to classify the shape of a breast mass.