Three‐dimensional reconstruction of cerebral vasculature from limited digital subtraction angiographic projections

An integrated approach is developed for three‐dimensional (3D) cone beam reconstruction of cerebral vasculature (both morphology and gray scale) from a limited number (4–6) of digital subtraction angiographic (DSA) projections, obtained using a standard biplane C‐arm x‐ray system, using the Algebraic Reconstruction Technique (ART) with non‐negativity constraint. The accuracy of ART is improved by determining weights based on the intersection volume between a pyramidal ray and cubic voxel. The input data are segmented and corrected for spatial image distortion. The reconstruction (volume ) is accelerated by retaining only the voxels containing vasculature (masking), multiscaling, and partitioning the algorithm to run in parallel. Problems in using ART are evaluated. Limited view reconstruction is demonstrated to be unstable. In addition, errors arising from geometric uncertainty, noise, and partial voluming are assessed with simulations. Corrective measures in reconstructions are described and demonstrated with experimental data. Reconstruction of a phantom containing fiducial markers at known locations demonstrated that the reconstructed geometry is accurate to less than a pixel width (0.23 mm). The reconstruction algorithm with corrective measures is tested using a thin wire phantom, and anatomic skull phantom, a human carotid artery, and human cerebral vasculature with an aneurysm. The use of weight norm cutoff and averaging ART for different terminal conditions is found to be effective. The reconstruction results are also validated with reprojections of the 3D reconstructions.