Computation and Storage Efficient Sparse MART Algorithm for 2-D, 3-D Reconstruction from Fan Beam, Cone-Beam Projection Data

Abstract

ABSTRACT Algebraic reconstruction algorithms are a better choice compared to transform-based algorithms whenever projection data is limited in nature. High computational cost and huge memory requirements are two major downsides of iterative reconstruction methods. Among all algebraic techniques, the Multiplicative Algebraic Reconstruction Technique (MART) is most popular because it maximizes the entropy (of the image) in the limiting case. In the present work, our ultimate goal is to reduce computational complexity and cope with the huge storage scenario of the MART algorithm. We propose a new sparse MART algorithm (Sp-MART) and test it with two-dimensional and three-dimensional (2D/3D) numerical data. A more accurate and efficient geometrical formula for calculating intersection length is also presented. Experimental projection data of human tooth and drip irrigation pipe is processed for further validation of the Sp-MART algorithm. Reconstructions of real specimens are also done using the FDK algorithm. The difference between two algorithms are investigated by calculating the structural similarity index (SSIM) and the L2 error of the results.