A Novel Active Contour Model for MRI Brain Segmentation used in Radiotherapy Treatment Planning

Abstract

INTRODUCTION Brain image segmentation is one of the most important clinical tools used in radiology and radiotherapy. But accurate segmentation is a very difficult task because these images mostly contain noise, inhomogeneities, and sometimes aberrations. The purpose of this study was to introduce a novel, locally statistical active contour model (ACM) for magnetic resonance image segmentation in the presence of intense inhomogeneity with the ability to determine the position of contour and energy diagram. METHODS A Gaussian distribution model with different means and variances was used for inhomogeneity, and a moving window was used to map the original image into another domain in which the intensity distributions of inhomogeneous objects were still Gaussian but were better separated. The means of the Gaussian distributions in the transformed domain can be adaptively estimated by multiplying a bias field by the original signal within the window. Then, a statistical energy function is defined for each local region. Also, to evaluate the performance of our method, experiments were conducted on MR images of the brain for segment tumors or normal tissue as visualization and energy functions. RESULTS In the proposed method, we were able to determine the size and position of the initial contour and to count iterations to have a better segmentation. The energy function for 20 to 430 iterations was calculated. The energy function was reduced by about 5 and 7% after 70 and 430 iterations, respectively. These results showed that, with increasing iterations, the energy function decreased, but it decreased faster during the early iterations, after which it decreased slowly. Also, this method enables us to stop the segmentation based on the threshold that we define for the energy equation. CONCLUSION An active contour model based on the energy function is a useful tool for medical image segmentation. The proposed method combined the information about neighboring pixels that belonged to the same class, thereby making it strong to separate the desired objects from the background.