Segmentation of mass in mammograms by a novel integrated active contour method

Abstract

Mass segmentation plays a vital role in the computer aided diagnosis systems for breast cancer. To efficiently detect the true boundaries of mass regions, a fully automated, dual-stage method was developed. Firstly an improved region-based level set method was applied for coarse segmentation and the gradient information was integrated to avoid boundary leaking. The obtained rough contour was used as an initial boundary for further refinement. In order to detect the delicate margins which are of significant meaning for breast diagnosis, a local geodesic active contour LGAC model based on local image information was proposed to refine the rough contour. The experimental results suggested that the proposed improved level set method can correctly find the radial and ambiguous edges of mass regions. Compared to the classical level set methods, the new scheme is more accurate and robust for mass segmentation in mammograms.