Breast mass segmentation based on information theory

Authors

Aize Cao, Nanyang Technological University, Singapore
Qing Song, Nanyang Technological University, Singapore
Xulei Yang, Nanyang Technological University, Singapore
Lei Wang, Nanyang Technological University, SingaporeFollow

RIS ID

54417

Publication Details

Cao, A., Song, Q., Yang, X. & Wang, L. (2004). Breast mass segmentation based on information theory. 17th International Conference on Patter Recognition (ICPR) (pp. 758-761). Australia: IEEE.

Abstract

In this study, an information-based algorithm, called c-shells based deterministic annealing (CSDA), is proposed for breast mass segmentation on digital mammograms. CSDA recasts the fuzzy clustering concept into the probability framework and offers two improved features over existing clustering algorithms. First, it is a global minimization algorithm through mass constrained deterministic annealing rather than a local minimization method in the original fuzzy c-shells (FCS) approach. Second, the prototype in this algorithm is shell, which is more effective in segmentation with compact or hollow spherical shells compared to the standard deterministic annealing (DA) algorithm. Experimental results show that the information based CSDA clustering algorithm is a promising image segmentation technique for digital mammographic mass detection.