3-DIMENSIONAL RECONSTRUCTION OF VASCULAR TREES - EXPERIMENTAL EVALUATION

This paper is the second of two that together present a novel approach to the problem of reconstructing vascular trees from a small number o f projections. Previously, we described the reconstruction algorithm a nd how it effectively circumvents the matching or ''correspondence pro blem'' found in most photogrammetric or computer-vision-based approach es. The algorithm is fully automatic and assumes that the imaging geom etry is known, the vascular tree is a connected structure, and that it s center-lines have been identified in three or more images. It employ s consistency and connectivity constraints and comprises three steps: The first generates a connected structure representing the multiplicit y of solutions that are consistent with the first two views; the secon d assigns a measure of agreement to each branch in this structure base d on one or more additional projections; and the third step employs th is measure to distinguish between those branches comprising the vascul ature and the accompanying artifacts. This paper addresses the issue o f validation via simulations and experiments. In addition to a clinica l case, we examine the performance of the algorithm when applied to si mulated projections of two 3-D vascular models, both representative of the complexity faced in coronary and cerebral angiography. The result s in each instance are impressive and demonstrate that adequate recons tructions may be obtained with as few as three distinct views. (C) 199 6 American Association of Physicists in Medicine.