Three-dimensional skeletonization for computer-assisted treatment planningin radiosurgery

This paper describes a new algorithm for skeletonization of two- (2D) and t hree-dimensional (3D) objects based on ridge extraction. Ridges are formed when grassfire fronts collapse during grassfire propagation and they corres pond to the locus of skeleton. The iso-distance contours/surfaces of a dist ance map are analogous to the grassfire fronts. They are locally smooth eve rywhere except at ridge locations. The new skeletonization algorithm extrac ts these ridge points based on local curvature measurement and is rotationa l invariant. It requires one scan of the image for curvature detection in a ny dimension, and is much faster than thinning methods. Connectivity checks are not required and the algorithm is extensible to higher dimensions. Our 3D skeletonization method is used in a novel algorithm to guide computeriz ed planning of radiosurgical treatment of brain tumors. (C) 2000 Elsevier S cience Ltd. All rights reserved.