Active Contour Models for Boundary Extraction with Application to Medical Images with Noise

Abstract

In medical image processing, active contour model is a method used to segment or extract the boundaries of an image for further processing. Recently, a selective active contour model called Selective Segmentation with Chessboard Distance (SSCD) model has been proposed to effectively segment a particular object in an image. However, the SSCD model has problems in extracting noisy images, which may result in poor segmentation. It is known that the presence of noise in some medical images cannot be avoided and can lead to poor segmentation. The aim of this research is therefore to reformulate the SSCD model to segment some medical images with noise. The modification is done by considering two different image denoising algorithms, the Gaussian filter and the bilateral filter, as new fitting terms in the SSCD model, resulting in two variants of modified SSCD models, referred to as SSCDG and SSCDB, respectively. The accuracy of the segmented image was evaluated using the Jaccard (JSC) and Dice similarity coefficient (DSC). Numerical experiments showed that the proposed SSCDG model based on the Gaussian filter denoising algorithm has the highest JSC and DSC values, which means the highest segmentation accuracy compared to the SSCD and SSCDB models. In the future, the proposed model can be extended to three-dimensional and color formulations.