Yes, seems that could work with some proper scripting.
What about the option in the FEAT GUI for "de-weighting" outliers?
Would it be possible to piggy-back into that, by having a mask for each
subject that explicitly defines what is and is not to be considered an
outlier? (i.e., set 0's to some small value 0.0001 so that they don't
automatically get masked but then create a "de-weighting mask" (and
weighting file?) that classifies those voxels in that subject as an
outlier that should be given 0 weight).
It isn't clear to me how to begin investigating this as a possible
option, because a search of the feat script for the terms "weight" and
"outlier" both come up empty.
thanks,
-MH
On Thu, 2010-06-24 at 18:13 +0000, Eugene Duff wrote:
>
> On 24 June 2010 17:48, Michael Harms <[log in to unmask]> wrote:
> Hi Eugene,
> Could you please elaborate briefly -- what sort of wrapper
> script are
> you thinking about to accomplish this?
>
>
>
> I haven't thought it right through..
>
>
> But you'd want to create the equivalent of the inputreg/maskunique
> file found in higher level analyses to generate masks for sets of
> voxels with equal numbers of inputs. For each of these sets of
> voxels, edit the original design file to run an analysis on only the
> data masked to keep those voxels. Combine the statistics and
> threshold.
>
>
> Something like that?
>
>
> Eugene
>
>
>
>
>
>
>
>
>
> thanks,
> -MH
>
> On Thu, 2010-06-24 at 16:15 +0000, Eugene Duff wrote:
>
> > Hi Michael -
> >
> >
> > FEAT is the same as SPM - a voxel with 0 in any of the
> inputs will be
> > excluded from the analysis. I'm not sure maps where the
> values of
> > different voxels reflect statistics associated with
> different subsets
> > of the total input set would usually be very interpretable.
> I doubt
> > other packages have the capability you're looking for. It
> shouldn't
> > be all that hard to write a wrapper script to FEAT that does
> what you
> > want though.
> >
> >
> > Cheers,
> >
> > Eugene
> >
> > --
> >
> > Centre for Functional MRI of the Brain (FMRIB) | University
> of Oxford
> > John Radcliffe Hospital | Headington
> > OX3 9DU | Oxford | UK
> >
> > Ph: +44 (0) 1865 222 523 | Mob: +44 (0) 7946 362 059 | Fax:
> +44 (0)
> > 1865 222 717
> >
> > --
> >
> >
> > On 24 June 2010 15:45, Michael Harms
> <[log in to unmask]> wrote:
> > Hello,
> > Continuing my fixation with voxels that are missing
> data
> > (i.e., voxels
> > with zeros), I was wondering how these voxels get
> treated
> > going to a 2nd
> > level analysis (or similarly, how voxels that
> weren't actually
> > acquired
> > at acquisition, due to limited slice coverage, get
> treated as
> > the first-
> > level images are transformed into standard-space for
> the 2nd
> > level
> > analysis).
> >
> > In particular, are such 0's treated as just any
> other
> > legitimate value,
> > or are they used to create a mask for the 2nd level
> images?
> > And, if the
> > latter, does a single instance of a 0 in a voxel of
> one
> > subject result
> > in a 0 at that voxel in the 2nd level results
> >
> > Basically, I'm trying to determine if there is any
> common
> > software
> > package for image statistics that can compute voxel-
> wise
> > statistics
> > using the subset of subjects that have data
> available at that
> > voxel in
> > standard-space, while just "ignoring" the subjects
> that don't
> > have data
> > at the voxel -- i.e., allowing for varying d.f's
> across
> > voxels in the
> > statistic computation. (For example, from what I
> can tell so
> > far, SPM
> > doesn't appear to have this capability, as a NaN at
> a voxel in
> > a single
> > subject results in a NaN at that voxel in the group
> level
> > contrast).
> >
> > I see that FEAT does have an option in Higher-level
> analysis
> > to
> > automatically "de-weight" outliers, which seems
> quite similar
> > in concept
> > (and perhaps could even be used to achieve the
> behavior that I
> > seek?)
> >
> > thanks,
> > -MH
> >
> >
> > --
> > Michael Harms, Ph.D.
> >
> --------------------------------------------------------------------
> > Conte Center for the Neuroscience of Mental
> Disorders
> > Washington University School of Medicine
> > Department of Psychiatry, Box 8134
> > Renard Hospital, Room 6604 Tel:
> 314-747-6173
> > 660 South Euclid Ave. Fax:
> 314-747-2182
> > St. Louis, MO 63110 Email:
> [log in to unmask]
> >
> --------------------------------------------------------------------
> >
> >
> >
>
>
>
|