Human Heart Segmentation Based on Differential Evolution and Active Contours with Shape Prior

Abstract

Active contour model is an image segmentation technique that uses the evaluation of internal and external forces to be attracted towards the edge of a target object. In this paper a novel image segmentation method based on differential evolution and active contours with shape prior is introduced. In the proposed method, the initial active contours have been generated through an alignment process of reference shape priors, and differential evolution is used to perform the segmentation task over a polar coordinate system. This method is applied in the segmentation of the human heart from datasets of Computed Tomography images. To assess the segmentation results compared to those outlined by experts and by different segmentation techniques, a set of similarity measures has been adopted. The experimental results suggest that by using differential evolution, the proposed method outperforms the classical active contour model and the interactive Tseng method in terms of efficiency and segmentation accuracy.