An Approach for Mammography Image Segmentation

This paper addresses the problem of segmenting tomographic images for the purpose of diagnosing breast cancer. In case of glandular and dense breasts cancers may be missed, due to the effect of overlapping tissues. The goal of this study is to develop, test, and validate an approach and algorithm for segmentation of breast masses from tomosynthesis images. The approach developed uses processing by elements for improving contrast of images using a specific function for extracting a preparation from the image. The image-preparation is segmented by a threshold criterion for uniformity and morphological operations. The threshold is determined based on the entropy of the image histogram. Developed algorithm is applied to real images with highly deteriorated contrast and shows stable results. An important characteristic of the approach is the ability of the result of segmentation to cover completely the sought find.