Pixel based reconstruction (PBR) techniques

Abstract

The authors have reinvestigated algebraic reconstruction algorithms for computed tomography, with some new heuristics and some new methodologies, which indicate that they merit serious consideration for practical applications. Methods of implementing these iterative methods more efficiently by reducing the computations, reducing memory requirement, and using accelerating techniques are discussed. It is shown by simulations, using the pixel-based reconstruction (PBR) methods proposed by R.S. Fager et al., that excellent reconstructions of a complex object can be obtained with very few projections. PBR methods are simultaneous in nature, i.e. at each iteration, any pixel value does not depend on any other pixel value in the reconstructed image. Because of this, in principle, each pixel can be assigned to its own computer and advantage can be taken of parallel structured computers. Two methods of improving these algorithms have been investigated: region of interest; and the steepest descent approach.<>