HeinzelCluster: accelerated reconstruction for FORE and OSEM3D

Author

Vollmar, S. ; Michel, C. ; Treffert, J.T. ; Newport, D.F. ; Casey, M. ; Knöss, C. ; Wienhard, K. ; Liu, X. ; Defrise, M. ; Heiss, W.-D.

Author_Institution

Max-Planck-Inst. fur Neurologische Forschung, Koln, Germany

Volume

3

fYear

2001

Firstpage

1766

Abstract

We have investigated two strategies to speedup reconstruction of data from our high resolution PET scanner (HRRT) on a dedicated cluster (7 nodes with 4 PIII Xeon@700 MHz, switched Fast Ethernet and Myrinet networking, Windows NT). With the first strategy, FORE/FOREJ and a fast 2D reconstruction method, we were able to reduce the 2D reconstruction time from 38 minutes to approx. 100 seconds. The second approach was to accelerate OSEM3D with a two-stage parallelization scheme resulting in a drop of reconstruction time from about 8 hours on one CPU to approx. 35 minutes for one core iteration.

Keywords

Fourier analysis; image reconstruction; local area networks; medical image processing; positron emission tomography; CPU; Fourier rebinning; HeinzelCluster; Myrinet networking; OSEM3D; Windows NT; cluster reconstruction; dedicated cluster; fast two dimensional reconstruction method; high resolution positron emission tomography scanner; one core iteration; switched Fast Ethernet networking; two dimensional reconstruction time; two-stage parallelization scheme; Acceleration; Ethernet networks; Libraries; Linux; Message passing; Nuclear medicine; Reconstruction algorithms; Scheduling algorithm; Software tools; Switches;

fLanguage

English

Publisher

ieee

Conference_Titel

Nuclear Science Symposium Conference Record, 2001 IEEE

ISSN

1082-3654

Print_ISBN

0-7803-7324-3

Type

conf

DOI

10.1109/NSSMIC.2001.1008684

Filename

1008684