Dear Elizabeth,
> We have a question concerning masking in a Random Effects model.
>
> We are interested in identifying activations that may be unique to the
> conceptual processing of a specific category of objects. To identify
> any unique regions of activation we must mask our original test
> contrast with both an inclusive and an exclusive mask.
>
> In our RFX analysis we have been using ImCalc to produce binary images
> of the contrasts that we want to use as masks. So for instance, the
> function i1>T (to specify the 'inclusive' mask) and (i1-1).*(-1) (to
> reverse this mask to produce an 'exclusive' mask). Where T = the T
> threshold for the mask.
>
> However, each time we apply a mask SPM calculates a small volume
> correction. So by the time we've applied both masks our search volume
> is greatly reduced. We end up with very significant results.
>
> As our design is not orthogonal we would like to be able to use masking
> in RFX in the same way as we do in FFX, i.e. without SVC. Is there any
> way to prevent SPM from using a small volume correction when we apply
> these masks?
Yes there is. After analysing your masked data at the second level use
the SVC (small volume correction) to specify a 'large' volume (equal in
size to the original search volume). SPM will compute corrected p
values on the basis that your search volume was the specified box or
sphere, even if the voxels encompassed were not actually analysed.
I hope this helps - Karl
|