Hi Maria,
It sounds as though these subjects just don't have any significant
voxels. The first warning you had concerned the first pass in which
SPM estimates non-sphericity by looking at the voxels which survive an
initial main effects F contrast -- with an uncorrected threshold of
0.001 (the original defaults.stats.fmri.ufp) no voxels survived this;
after you set it to 0.05 some voxels do.
Presumably later in the results step you are specifying a more
stringent level than 0.05 uncorrected (e.g. maybe you are using FWE)
and again, nothing survives this level. You could reduce your results
threshold, but I'd worry slightly that t=1.3 is already very low, any
lower might not really be meaningful.
So, what now? Well, one answer is that maybe these subjects really
don't have any significant voxels, that doesn't cause an error (just a
warning from SPM) and as far as I am aware, there is no law that says
an fMRI experiment must find something significant!
Another answer, is that you can go ahead and do a second level
analysis of the group of subjects. The "con" images will be analysed,
and these will be correct, regardless of whether their corresponding t
images contain significant voxels.
However, it's also possible that the lack of significance stems from a
problem with the acquisition or preprocessing of the scans for these
subjects. E.g. maybe there is a large artefact, or some kind of
failure in the realignment, which causes large voxelwise variance, and
hence low t-values. So you should go through and check your original
images, and the intermediate steps in your preprocessing. You might
also find one of the automatic data-checking methods useful, e.g.
http://www.sph.umich.edu/~nichols/SPMd/
http://gablab.stanford.edu/tools.htm (maybe not SPM5 compatible)
(probably others...)
Best,
Ged.
<Maria> <Coletta> wrote:
> Hi,
>
> On Sunday, I posted a question about the Estimate step (using SPM5, under
> Matlab 6.5, working with a data set of 26 subjects. In the "Estimate"
> step, for about half my subjects I get this warning dialog popup- "Please
> check your data, there are no significant voxels. The globals are plotted
> for diagnosis.")
>
> I found and followed this suggestion from the archives:
> set 'defaults.stats.ufp' to be less stringent, eg .05, which may allow
> your model to run ok, even though the first pass is irrelevant in one-
> sample t-tests (ie the results should be unaffected, since there should be
> no weighting required, ie SPM.xX.W should be estimated as speye(n)
> anyway?).
>
> This did allow me to complete estimation for those subjects, however now
> for these subjects, when I go to set the t-contrasts (design is attached)-
> I get this error:
>
> SPM computation : ...initialising
> SPM computation : ...done
> Warning: No voxels survive height threshold u=1.3
>> In c:\MATLAB6p5\spm5\spm_getSPM.m at line 631
> In c:\MATLAB6p5\spm5\spm_results_ui.m at line 264
> SPM computation : ...done
>
> Does anyone have any suggestions? I'm at a loss!
>
> Thanks in advance!
|