Advanced search
1 file | 246.65 KB

Advanced tomography reconstruction algorithms on the graphical processing unit

Yoni De Witte (UGent) , Jelle Vlassenbroeck (UGent) , Manuel Dierick (UGent) and Luc Van Hoorebeke (UGent)
Author
Organization
Abstract
For many years the FDK algorithm has been the standard algorithm for the reconstruction of projection data acquired in cone beam geometry, which is the standard geometry in high resolution X‐ray tomography. Iterative reconstruction algorithms such as SART [1] can however provide superior image quality. Additionally, artifacts associated with the cone beam geometry can be overcome by using alternative scanning trajectories such as the helical cone beam geometry, of which the data can for instance be reconstructed using the Katsevich algorithm [2]. However, the advanced algorithms for iterative reconstruction and helical cone beam reconstruction are far more complex than the FDK algorithm. The combination of this computational complexity and the large datasets encountered in high resolution X‐ray tomography has prevented their application in this field. The recent breakthrough of ‘manycore’ computing devices such as the Graphical Processing Units now allows the practical use of both helical cone beam geometry and iterative reconstruction methods in high resolution X‐ray tomography. Here we present the implementation of these algorithms on the NVDIA CUDA architecture in the Octopus reconstruction package [3]. The performance of the algorithms is illustrated by means of timing results for typical data volumes. Additionally, the advantages with respect to image quality and the reduction of cone artifacts are illustrated by presenting reconstruction results for several applications (e.g. figures 1 and 2).
Keywords
X-ray microscopy, GPU, algorithm

Downloads

  • (...).pdf
    • full text
    • |
    • UGent only
    • |
    • PDF
    • |
    • 246.65 KB

Citation

Please use this url to cite or link to this publication:

Chicago
De Witte, Yoni, Jelle Vlassenbroeck, Manuel Dierick, and Luc Van Hoorebeke. 2010. “Advanced Tomography Reconstruction Algorithms on the Graphical Processing Unit.” In XRM2010 : 10th International Conference on X‐Ray Microscopy, Book of Abstracts, 226–226. Chicago, IL, USA: Argonne National Laboratory.
APA
De Witte, Y., Vlassenbroeck, J., Dierick, M., & Van Hoorebeke, L. (2010). Advanced tomography reconstruction algorithms on the graphical processing unit. XRM2010 : 10th International conference on X‐Ray Microscopy, Book of abstracts (pp. 226–226). Presented at the 10th International conference on X‐Ray Microscopy (XRM 2010), Chicago, IL, USA: Argonne National Laboratory.
Vancouver
1.
De Witte Y, Vlassenbroeck J, Dierick M, Van Hoorebeke L. Advanced tomography reconstruction algorithms on the graphical processing unit. XRM2010 : 10th International conference on X‐Ray Microscopy, Book of abstracts. Chicago, IL, USA: Argonne National Laboratory; 2010. p. 226–226.
MLA
De Witte, Yoni, Jelle Vlassenbroeck, Manuel Dierick, et al. “Advanced Tomography Reconstruction Algorithms on the Graphical Processing Unit.” XRM2010 : 10th International Conference on X‐Ray Microscopy, Book of Abstracts. Chicago, IL, USA: Argonne National Laboratory, 2010. 226–226. Print.
@inproceedings{1105930,
  abstract     = {For many years the FDK algorithm has been the standard algorithm for the reconstruction of projection data acquired in cone beam geometry, which is the standard geometry in high resolution X\unmatched{2010}ray tomography. Iterative reconstruction algorithms such as SART [1] can however provide superior image quality. Additionally, artifacts associated with the cone beam geometry can be overcome by using alternative scanning trajectories such as the helical cone beam geometry, of which the data can for instance be reconstructed using the Katsevich algorithm [2]. However, the advanced algorithms for iterative reconstruction and helical cone beam reconstruction are far more complex than the FDK algorithm. The combination of this computational complexity and the large datasets encountered in high resolution X\unmatched{2010}ray tomography has prevented their application in this field. The recent breakthrough of {\textquoteleft}manycore{\textquoteright} computing devices such as the Graphical Processing Units now allows the practical use of both helical cone beam geometry and iterative reconstruction methods in high resolution X\unmatched{2010}ray tomography. Here we present the implementation of these algorithms on the NVDIA CUDA architecture in the Octopus reconstruction package [3]. The performance of the algorithms is illustrated by means of timing results for typical data volumes. Additionally, the advantages with respect to image quality and the reduction of cone artifacts are illustrated by presenting reconstruction results for several applications (e.g. figures 1 and 2).},
  author       = {De Witte, Yoni and Vlassenbroeck, Jelle and Dierick, Manuel and Van Hoorebeke, Luc},
  booktitle    = {XRM2010 : 10th International conference on X\unmatched{2010}Ray Microscopy, Book of abstracts},
  keyword      = {X-ray microscopy,GPU,algorithm},
  language     = {eng},
  location     = {Chicago, IL, USA},
  pages        = {226--226},
  publisher    = {Argonne National Laboratory},
  title        = {Advanced tomography reconstruction algorithms on the graphical processing unit},
  year         = {2010},
}