An Algorithm for Suggesting Delineation Planes

In this paper we present a scheme to reduce the amount of user iterations required to segment an object by delineating on cross-section planes. Starting with an initial segmentation created from a small number of delineated curves, the algorithm progressively analyzes the uncertainty of segmentation with respect to the image features and suggests the “next plane” for delineation that would maximally resolve the uncertain regions. Compared with the few prior art on this problem, we adopt a simpler computational framework that is made up of an RBF-based curve interpolation method, a distance-based uncertainty metric, and a plane determination approach using density-based clustering. We demonstrate using both synthetic and real examples that our method uses less than 50% of the number planes than a random selection scheme to achieve 90% segmentation accuracy.