Hi - yes, the stats are computed separately for each voxel so it is fine
to do this kind of parallelisation. Because MCMC (which includes the use
of random processes) is used in full FLAME there is a _very_ small change
in the final z stat each time it is run - in the latest version of FSL
this is tuned for high accuracy and repeatability (including for example
by running the MCMC for long enough which is why FLAME takes a while to
run - MCMC is generally thought of as "gold-standard but slow") so that
the variation in z is generally less than around 0.01.
Thanks, Steve.
On Mon, 29 Dec 2003, KORB,ALEX S wrote:
> I've noticed that if there are more than 40 subjects in a group analysis
> then FLAME computes the statistics slice by slice. I assume this means
> that the statistics are slice independent. Under this assumption I've been
> trying to parallelize the analysis on a computer cluster. However, the
> zstats calculated when FLAME is run by slice vs regular FLAME are slightly
> different. Anyone know why this is? Thanks, Alex Korb
>
Stephen M. Smith DPhil
Associate Director, FMRIB and Analysis Research Coordinator
Oxford University Centre for Functional MRI of the Brain
John Radcliffe Hospital, Headington, Oxford OX3 9DU, UK
+44 (0) 1865 222726 (fax 222717)
[log in to unmask] http://www.fmrib.ox.ac.uk/~steve
|