Multiscale segmentation and enhancement in mammograms

Abstract

A multiscale method for segmenting and enhancing lesions of various sizes in mammograms is presented. The method uses two stages. The first stage applies a multiscale automatic threshold estimator based on histogram moments to segment the mammogram at multilevels. The second stage converts the segmented image using pseudo-colour mapping to produce a colour image. An algorithm is presented as well as experimental results. Mammograms are digitalised using a table scanner with a transparency adapter and the algorithm is implemented in Borland C++ 4.02 on a 486 PC. The results are analogous to a breast map which provide an adequate basis for radiological breast tissue differentiation and analysis in digital mammography. Experimental results and judgments from radiological experts are very encouraging.