Optimal graph-theoretic approach to 3D anatomical tree matching

High resolution three-dimensional (3D) images are readily produced by many imaging scanners. This paper describes a novel optimal method for matching the anatomical trees contained in such 3D images. Our method explicitly describes a set of valid candidate matches between two input trees by considering topological deformations known to occur during the tree definition process, evaluates these candidate matches using a cost function that compares corresponding branch and branchpoint attributes measured from the 3D image data, and locates a globally optimal match with respect to the cost function using an efficient dynamic programming algorithm. We present matching results for human airway trees. The method is a part of a complete computer-based system for 3D tree analysis