Statistical Parametric Mapping
The Statistical Parametric Mapping (SPM) package was developed by members and collaborators of the
Wellcome Department of Cognitive Neurology. Preliminary work has been done at the University of
Tennessee in Knoxville to improve the performance of SPM by porting aspects of the package to NetSolve.
The following summary of SPM's focus is from the SPM web site:
Statistical Parametric Mapping refers to the construction and assessment of spatially extended
statistical process used to test hypotheses about [neuro]imaging data from SPECT/PET & fMRI.
These ideas have been instantiated in software that is called SPM.
NetSolve and SPM
Although parts of the SPM code are inherently parallel, very little work has been done previously
to take advantage of this program structure. Recent work has been done at the University of Tennessee
to exploit some of SPM's parallelization through NetSolve. As an example, when aligning a series of
picture slices, much of the computation being performed on each slice is completely independent from
other slices. NetSolve's task farming interface provides a natural means for sharing work across
several NetSolve servers.
SPM requires NetSolve's Matlab interface in order to carry out its parallelization. Through a series
of Matlab scripts and Matlab compiled executables using NetSolve's Matlab interface, executions have
been timed to run in one half to one third the time of unmodified code in some simulations. Work on
integrating SPM and NetSolve is still in early stages and is not currently ready for use.
|