Adaptive local-fitting-based active contour model for medical image segmentation

Abstract

Due to the complex environments, characteristics of samples and limits of acquisition instruments, medical images are often acquired with intensity inhomogeneity and noise. These two consequences have brought challenges in medical image segmentation, especially when the background has similar intensity with the region of interest (ROI). Although lots of active-contour-model-based methods have appeared to extract ROI accurately, finer segmentation especially in tiny areas is still a difficult problem. In this paper, an adaptive local-fitting-based active contour model is proposed to separate the ROI accurately and robustly, which takes both adaptive local fitting energy and regularization energy to drive the initial contour to the object boundary. Unlike the traditional methods which assume the intensities in local region are constant, the proposed method takes an adaptive technique to fit the original image, which contributes to a more optimal solution. Experiments are conducted on lots of synthetic and real medical images. Results show that the proposed method not only obtains more accurate segmentation results than several well-known algorithms, while is robust to intensity inhomogeneity and noises. The source code is available at: https://github.com/madd2014/ALF.