Aha - Joe was right - sorry Joe!
You are right in saying that the maths in FLAME would automatically
correct for reduced DOF (etc) in voxels feeding from a reduced number of
subjects. However, in order to process these peripheral voxels where not
all subjects are present would need some rewriting inside FLAME; you're
right, it's not currently an option. Of course, if you are particularly
concerned about a particular voxels, you can always use avwroi to extract
that from the relevant subject's copes and varcopes and call flame (see
the report.com file to get the syntax) but this would be rather painful
for you!
A plausible solution though is this: after all the first-level FEAT
directories have had "applyfeatreg" run on them (by the first time you ran
a higher-level analysis) you should run the following on every image in
each firstlevel.feat/reg_standard image (and the subdirectories in there);
avwmaths <input> -dil -dil <input>
(I suggest you script this as there will be lots of these images!)
this will simply dilate the data, so that when you run group-stats you
will be less likely to have voxels with some subjects missing. The
original voxels (with all subjects) will have the same results, but the
voxels you are worried about should now get processed as well (not with
the reduced number of subjects, but, for the subjects that were missing
before, with the data from the nearest valid voxels; this isn't too dodgy,
as you are only allowing a movement of 4mm, and subject spatial
variability is probably that big anyway). Good luck!
Thanks, Steve.
On Mon, 24 Mar 2003, Edward Vessel wrote:
> Actually, my question was quite different altogether =)
>
> Let me try rephrasing:
>
> My subject is a complete within-subjects design. So, I could get an estimate
> of my contrasts of interest from any subset of my 16 subjects. The best
> would be to use all 16 subjects for every voxel. However, since I can only
> image a limited space (16 slices per subject), the realignment of my data
> space (after stripping of non brain tissues) will mean that some voxels only
> actually have data (but in all conditions) from a subset of my subjects.
>
> The default behavior of feat seems to be to AND together all the individual
> masks, which for me means I only get group data on the reduced volume that
> was common to each subject's data space. It would be nice, however, if I
> could still get an estimate from the other voxels which might be missing only
> one, or two, or three subjects ... if they were weighted by the number of
> subjects contributing, then this should be able to adjust the variance of
> each voxel (by reducing the DOF). Therefore, a voxel with only a single
> subject contributing data might be virtually ignored (due to the huge
> variance and resulting miniscule t value), while a voxel with 15 subjects
> contributing might still produce decent data.
>
> The problem with just changing the individual subject masks is that then I am
> entering an 'empty' voxel with zero variance into the analysis ... how this
> would effect the flame computations depends obviously on how the program was
> written.
>
> Ed
>
>
> Stephen Smith wrote:
>
> >yes, I think you are; the question is just asking how to restrict the
> >voxels that FLAME runs on, instead of the default (of all voxels which are
> >present in all subjects); ie how to setup a ROI analysis pre-stats.
>
> :)
>
>
> On Mon, 24 Mar 2003, Joe Devlin wrote:
>
> > > Then go into there and I _THINK_ that if you edit the mask volume in there
> > > then that will affect everything else in a higher-level analysis.
> >
> > This does raise the question as to why you would want to include voxels
> > without any values in your analysis though? If a voxel has a zero value
> > in one (or more) subjects and this is analysed along with voxels
> > containing measurements from the rest of the subjects, it wouldn't seem to
> > yield a meaningful result. Mostly it seems to tell you that some
> > pre-processing step along the line didn't work so well and things aren't
> > sufficiently aligned (at that voxel at least).
> >
> > Perhaps I misunderstood the question?
> > Cheers,
> > Joe
>
>
> On Monday 24 March 2003 04:34 am, Stephen Smith wrote:
> > Hi - you will need to:
> >
> > Take (any) one of the first-level sessions and make sure that it has the
> > subdirectory reg_standard complete (this is the stats in standard space
> > which gets created the first time you run higher-level FEAT). If not you
> > can force this creation with
> > featregapply <featdirname>.feat
> >
> > Then go into there and I _THINK_ that if you edit the mask volume in there
> > then that will affect everything else in a higher-level analysis.
> >
> > thanks, Steve.
> >
> > On Mon, 24 Mar 2003, Edward Vessel wrote:
> > > hi folks -
> > >
> > > I have b
>
> --
> Ed Vessel
> U. of Southern California [log in to unmask]
> Dept. of Neuroscience
> HNB, 3641 Watt Way http://geon.usc.edu/~vessel
> Los Angeles, CA 90089-2520
> (213) 740-6102
>
Stephen M. Smith MA DPhil CEng MIEE
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
|