An Intrinsic Image Representation and Its Application to Left Ventricle Segmentation in Cardiac MRI Images

Abstract

In this work, we presents an interactive image segmentation algorithm based on the extraction of isocontours at specified points from an intrinsic image representation that is obtained by our image decomposition algorithm. Our image decomposition is designed to remove undesirable bias field and noise from a variety of medical images in a variational framework and our segmentation is developed using isocontours obtained by a marching cube algorithm via interactive selection of initial points of the contours on the intrinsic image where the effect of undesirable bias field and noise are reduced. The experimental results demonstrate the robustness and effectiveness of our algorithm in the delineation of left ventricles in cardiac MRI images, and show a potential as the basis for an image annotation tool in the diagnosis or treatment procedure based on medical images.