M. Xu and P. Thulasiraman (Canada)
Computed Tomography, Iterative Reconstruction Technique, Ordered Subset Simultaneous Algebraic Reconstruction Technique, Siddon’s Algorithm, Rotation-based Projector and Backprojector
Iterative reconstruction techniques are safe approaches for computed tomography (CT). However, the long process ing time of them prevent their adoption in clinical CT machines. Recently, with the growing increase in computa tional power due to multiple cores on a chip, iterative techniques are no longer deemed unsolvable. In this paper, we investigate one category of iterative techniques which stem from algebraic reconstruction technique, including ART (algebraic reconstruction technique), SIRT (simultaneous iterative reconstruction technique), SART (simultaneous algebraic reconstruction technique), and ordered subset version of SART (OS-SART). We exploit the parallelism in OS-SART and its feasibility to parallelization on homogenous shared memory multicore architectures. OS SART is memory bound due to the frequent access to a large weighting factor matrix. Therefore, to reduce memory latency in OS-SART, we parallelize OS-SART algorithm by incorporating rotation-based projector and back projector which enables the computation of weighting factors at runtime on a need basis. The results show that our parallel algorithm is scalable on homogeneous multicore architecture in terms of both core number and problem size. For 8 cores, the speedup increases from 6.77 for image size of 128x128 to 7.02 for image size of 512x512. For 16 cores, the speedup increases from 12.29 for image size 128x128 to 13.18 for image size 512x512.
Important Links:
Go Back
