Gaussian Pyramid Based Multi-Scale GVF Snake for Mass Segmentation in Digitized Mammograms

Abstract

Mass segmentation plays an important role in many computer-aided diagnosis (CAD) system. It is usually used as the previous step of mass classification. In this paper, we propose one novel scheme for segmentation of breast mass in digitized mammograms, which is based on gradient vector flow (GVF) snake and multi-scale analysis using Gaussian pyramid. In the proposed method, mammogram is decomposed by Gaussian pyramid into a sequence of images from fine to coarse firstly, at the large scale, the image is coarse without much noise and artifact, so the GVF snake is able to converge to the mass contour easily and quickly without too much computation. Then the obtained contour is mapped to a higher resolution and used as the initial position to get a higher resolution result with GVF snake. Repeat this process until a correct mass boundary is attained. In addition, a parameter based on region variation is used for GVF snake as a stop criteria of the iterations and it reduces the number of unnecessary iterations. The experimental results demonstrate that the proposed method is more efficient and robust than the traditional GVF snake.