Dear Nissen,
> I have 10 set of fmri data collected in terms of A-R-B-R-C-R-D-R
> design. The four activation tasks, A, B, C, and D were interleaved with
> a common baseline of fixation condition, R. To compare each activation
> task to fixation condition, I can get a very significant level
> (corrected p< 0.05) and widespread activation maps. However the
> difference between the four activation tasks is very subtle, activation
> level and region of direct comparisons of the four tasks were reduced
> to a very large extent. Among the possible contrasts I am interested
> in activation difference of some planned contrasts only, e.g. (A-B),
> (B-C), and (B-D). However activation foci revealed by any planned
> contrast may represent the difference of the deactivation of the two
> conditons. In order to avoid this, each planned contrast was masking
> with their main contrast, e.g. mask contrast (A-B) with contrast (A-R)
> inclusively, and the p threshold of the mask contrast was set at p=
> 0.05 uncorrected (default setting). Is this operation reasonable?
Yes it is. It is a selective but principled reporting of significant
voxels.
> If yes, how is the p threshold for the mask contrast? Is it necessary to
> be set at a very high level (e.g. p=0.05 corrected) ?
The threshold for the mask can be any value. The masking does not
enter into the (height-based) inference about the A-B contrast, it just
selects voxels on the basis of another criteria. p<0.05 seems fine if
you simply wish to exclude regions that deactivated during A.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|