A foreground marker based centroid initialized Geodesic active contours for histopathological image segmentation

Abstract

Nuclear segmentation is considered to be one of the major challenge in the field of Histopathological Imaging. Various segmentation approaches have been proposed in the literature. The quality of the histopathological images have posed various challenges to those proposed techniques and they all suffer with deficiencies due to poor edge information and irregularities of the boundary. Active contours are considered to be the promising solutions to such a challenging task. The major issues with Active contours are computation of gradient information, initialization and occlusion detection. To address these issues effectively, an edge gradient driven Geodesic active contour(GAC) with a novel approach of detecting seed points based on foreground markers is proposed in this paper. The experimentation is performed on breast cancer tissue images and the efficiency measures such object detection accuracy and overlap resolution have been computed and compared with that of GAC without foreground markers as referred to the ground truth opined by the pathologists from Department of Pathology, JSS Hospital.