Fast iterative reconstructions for animal CT

H. M. Huang, I. T. Hsiao*, M. L. Jan

*Corresponding author for this work

Research output: Contribution to journalJournal Article peer-review

Abstract

For iterative x-ray computed tomography (CT) reconstruction, the convex algorithm combined with ordered subset (OSC) [1] is a relatively fast algorithm and has shown its potential for low-dose situations. But it needs one forward projection and two backprojections per iteration. Unlike convex algorithm, the gradient algorithm only requires one forward projection and one backprojection per iteration. Here, we applied ordered subsets of projection data to a modified gradient algorithm. In order to further reduce computation time, the new algorithm, the ordered subset gradient (OSG) algorithm, can be adjusted with a step size. We also implemented another OS-type algorithm called OSTR. The OSG algorithm is compared with OSC algorithm and OSTR algorithm using three-dimensional simulated helical cone-beam CT data. The performance is evaluated in terms of log-likelihood, contrast recovery, and bias-variance studies. Results show that images of OSG has compatible visual image quality to those of OSC and OSTR, but in the resolution and bias-variance studies, OSG seems to reach stable values with faster speed. In particular, OSTR has better recovery in a smoother region, but both OSG and OSC have better recovery in the high-frequency regions. Moreover, in terms of log likelihood with respect to computation time, OSG has faster convergence rate than that of OSC and similar to that of OSTR. We conclude that OSG has potential to provide comparable image quality and is more computationally efficient, and thus could be suitable for low-dose, helical cone-beam CT image reconstruction.

Original languageEnglish
Article numberP06017
JournalJournal of Instrumentation
Volume4
Issue number6
DOIs
StatePublished - 2009

Keywords

  • Data processing methods
  • Image reconstruction in medical imaging

Fingerprint

Dive into the research topics of 'Fast iterative reconstructions for animal CT'. Together they form a unique fingerprint.

Cite this