Medical Image Segmentation Using Modified Level-Set Model with Multi-Scale Gradient* Vector Flow

Abstract

This paper presents a novel method for medical image segmentation that can detect the edges or boundaries of all target objects(defined as high intensity regions) in an image by integrating multi-scale gradient* vector flow(MGVF) into a modified level-set model. The MGVF uses multi-scale images and the gradient of gradient magnitude of a scaled image to generate a vector flow field. This vector flow field is then substituted into a corresponding partial differential equation(PDE) of a modified level-set model that represents the active contour. The proposed method can effectively pull the active contour to attach to the boundary of each target object in an image, especially the boundary of an object that is very close to another object and the boundary of an object with low gradient magnitude. The experiments were tested on 1600 two dimensional CT scan images and the results have shown that the proposed method can accurately detect the boundaries of bones, colons, and residuals inside the colons.