Hi Cédric,
> I agree that the original time-series should not be affected by the
> choice of the contrast but indeed they are.
>
> Actually, the Y values extracted via the SPM VOI button seem to be
> affected by the chosen contrast (even if they are not adjusted)
I think this might be what's going on:
- spm_spm sets SPM.xVol.XYZ to include *all in-mask* voxels
- spm_results_ui calls spm_getSPM
- spm_getSPM then performs thresholding for the specified contrast
- spm_getSPM sets xSPM.XYZ to include *only supra-threshold* voxels
- spm_results_ui then passes both SPM and xSPM to spm_regions
- spm_regions then uses xSPM.XYZ, see e.g. line 182:
y = spm_get_data(SPM.xY.VY,xSPM.XYZ(:,Q));
So the actual voxels for which the eigenvariate are computed and
returned depend on what survived the chosen contrast, and hence differ
even if their values are not subsequently adjusted for the contrast's
null space.
If I've got this right, I'm not sure whether it's desirable behaviour
or not -- like you, I guess I would have expected the voxels to depend
on the specified VOI and not on the chosen contrast or threshold
levels. Maybe the idea is that the VOI should not be diluted with
unactivated voxels? Though I would guess that motivation would depend
on the purpose for which you were extracting the VOI. Or maybe we are
both missing another reason why it is done like this...
It seems to me that spm_regions could be fairly easily modified to use
SPM.xVol.XYZ instead of xSPM.XYZ (and similarly for XYZmm) if you are
certain you want to include sub-threshold voxels, but I haven't tried
this...
Incidentally, I found this webpage
http://airto.bmap.ucla.edu/BMCweb/HowTo/time.html
which mentions the eigenvariate extraction, and seems to think it's
peculiar. However, they seem to believe the dependence on the "model"
(contrast?) comes from the overall effects of interest F-test, which I
don't agree with, as this occurs in spm_spm, and doesn't appear to be
used to mask the voxels present in SPM.xVol.XYZ.
I hope this helps, and that someone else can shed some more light on
the decision to only look at the intersection of the specified VOI and
the supra-threshold regions of the specified contrast.
Best,
Ged
|