gms | German Medical Science

Kongress Medizin und Gesellschaft 2007

17. bis 21.09.2007, Augsburg

Medigrid: distributed computing to accelerate fMRI analysis

Meeting Abstract

  • Michael Luchtmann - Medical Faculty, Institute for Biometry and Medical Informatics, University of Magdeburg, Magdeburg
  • Sebastian Baecke - Medical Faculty, Institute for Biometry and Medical Informatics, University of Magdeburg, Magdeburg
  • Ralf Lützkendorf - Medical Faculty, Institute for Biometry and Medical Informatics, University of Magdeburg, Magdeburg
  • Johannes Bernarding - Medical Faculty, Institute for Biometry and Medical Informatics, University of Magdeburg, Magdeburg

Kongress Medizin und Gesellschaft 2007. Augsburg, 17.-21.09.2007. Düsseldorf: German Medical Science GMS Publishing House; 2007. Doc07gmds247

Die elektronische Version dieses Artikels ist vollständig und ist verfügbar unter: http://www.egms.de/de/meetings/gmds2007/07gmds247.shtml

Veröffentlicht: 6. September 2007

© 2007 Luchtmann et al.
Dieser Artikel ist ein Open Access-Artikel und steht unter den Creative Commons Lizenzbedingungen (http://creativecommons.org/licenses/by-nc-nd/3.0/deed.de). Er darf vervielfältigt, verbreitet und öffentlich zugänglich gemacht werden, vorausgesetzt dass Autor und Quelle genannt werden.


Gliederung

Text

Introduction: Preprocessing fMRI data (compensating motion artefacts by realigning the images, normalizing them to a standard brain template, and smoothing them to enhance signal-to-noise ratio) is very time consuming. Parallel architectures such as clusters or GRIDs can speed up pre-processing considerably. Within the framework of the German MediGRID we used up to 4 computers to decrease overall pre-processing time by a factor of 3.

Methods: The concept was realized by distributing selected image subsets and one identical reference image on different computers of the cluster. The subsets were independently pre-processed. Subsequently, the pre-processed subsets were merged together for the final statistical analysis that is based on a public domain software (SPM) using a General Linear Model. We used up to 4 PC (Windows XP, 2 GB RAM, Intel Pentium D) where MATLAB 7 and spm2 was installed. A user interface was developed to distribute and merge the images on the GRID, to coordinate the distributed resources, and to start the local MATLAB batch scripts via DCOM [1].

Results: Benchmarks for the evaluation of 50 to 400 fMRI data (64x64x32 voxel) showed an almost linear increase of the data evaluation speed with the number of computers. Only two groups reported results concerning parallelized spm using MPI [2] or CORBA [3]. A comparison with our results is difficult as both groups did not present benchmarks.

Conclusion: As matlab can also be run on unix platforms our concept can be easily implemented on heterogeneous platforms underlying most GRIDs. It requires only the implementation of the widely used MATLAB and the public-domain fMRI analysis software SPM while MIP and CORBA might not be available on general GRID architectures. Our solution can also be easily extended to larger clusters as well as to apply other methods such as Web services or RPCs.

Referenzen:

[1] http://www.mathworks.com/access/helpdesk/help/techdoc/matlab external/f64299.html, 2006.

[2] Jejo Koola. ParallelizedSPM2. http://www.fil.ion.ucl.ac.uk, 2006.

[3] Marcel May; Frank Munz; Thomas Ludwig. CORBA-basierte verteilte Berechnung medizinischer Bilddaten mit SPM. Bildverarbeitung für die Medizin, pp 213–217, 2000.


References

1.
http://www.mathworks.com/access/helpdesk/help/techdoc/matlab external/f64299.html, 2006. Externer Link
2.
Jejo Koola. ParallelizedSPM2. http://www.fil.ion.ucl.ac.uk, 2006 Externer Link
3.
Marcel May, Frank Munz, Thomas Ludwig. CORBA-basierte verteilte Berechnung medizinischer Bilddaten mit SPM. Bildverarbeitung für die Medizin, pp 213–217, 2000