Share Email Print

Proceedings Paper

GPU-based iterative reconstruction with total variation minimization for micro-CT
Author(s): S. M. Johnston; G. A. Johnson; C. T. Badea
Format Member Price Non-Member Price
PDF $17.00 $21.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

Dynamic imaging with micro-CT often produces poorly-distributed sets of projections, and reconstructions of this data with filtered backprojection algorithms (FBP) may be affected by artifacts. Iterative reconstruction algorithms and total variation (TV) denoising are promising alternatives to FBP, but may require running times that are frustratingly long. This obstacle can be overcome by implementing reconstruction algorithms on graphics processing units (GPU). This paper presents an implementation of a family of iterative reconstruction algorithms with TV denoising on a GPU, and a series of tests to optimize and compare the ability of different algorithms to reduce artifacts. The mathematical and computational details of the implementation are explored. The performance, measured by the accuracy of the reconstruction versus the running time, is assessed in simulations with a virtual phantom and in an in vivo scan of a mouse. We conclude that the simultaneous algebraic reconstruction technique with TV minimization (SART-TV) is a time-effective reconstruction algorithm for producing reconstructions with fewer artifacts than FBP.

Paper Details

Date Published: 22 March 2010
PDF: 11 pages
Proc. SPIE 7622, Medical Imaging 2010: Physics of Medical Imaging, 762238 (22 March 2010); doi: 10.1117/12.844368
Show Author Affiliations
S. M. Johnston, Duke Univ. Medical Ctr. (United States)
G. A. Johnson, Duke Univ. Medical Ctr. (United States)
C. T. Badea, Duke Univ. Medical Ctr. (United States)

Published in SPIE Proceedings Vol. 7622:
Medical Imaging 2010: Physics of Medical Imaging
Ehsan Samei; Norbert J. Pelc, Editor(s)

© SPIE. Terms of Use
Back to Top