Binary vascular reconstruction from a limited number of cone beam projections.

This paper describes a method to perform reconstruction of vascular cross-sectional images from a limited number of x-ray angiographic cone-beam projections. It is assumed that the projection data can be simplified by identifying blood vessels in each angiogram and removing signals due to other structures. Under these conditions, the x-ray attenuation coefficient, mu, can be modeled as a binary variable having a value mu 0 within the vessel and "0" outside. The reconstruction is performed by minimizing a cost function using the method of simulated annealing. In this paper, we demonstrate that the introduction of a priori information allows one to reconstruct a sphere and a simulated branched vessel from three views with, respectively, 97% and 93% of voxels having correct values. The addition of a continuity constraint for the reconstruction of the branched vessel resulted in further reduction in the percentage of misplaced voxels. Calculations require from one to six hours of CPU time on a Sun SparcStation 2 computer for the cases investigated here. The effect of noise, "cooling" schedule, and number of views on the reconstruction are examined using simulated vessel projections. Modifications to our approach to accelerate the reconstruction are also discussed.