Victor,
> I am doing a non-parametric second-level analysis (12 con images) with
> SnPM-99.
>
> I understand that there is no "SVC" option in SnPM, however could I
> still anyhow benefit by having a spatially-constrained hypothesis
> about activation?
Definitely! Anytime you can constrain your analysis you can generally
increase your sensitivity. Specifically, by using a ROI you are
reducing the severity of your multiple comparisons problem, and hence
will be able to use a lower threshold, and so get more power.
> For the sake of experiment I tried to perform an SnPM99 analysis on
> con-images previously masked by a VOI. I noticed that the "critical
> threshold" reported at the bottom of Table is lower for the VOI
> analysis then for the whole brain analysis, both were done with the
> same variance smoothing, number of permutations, thresholds (voxel
> p<0.05, no extent threshold), etc. Moreover the VOI-based analysis
> revealed a significant activation that did not show up during the
> whole-brain analysis.
Yup... this is totally consistent with what I wrote above. The VOI is
yielding a lower threshold and hence greater sensitivity. Makes
sense, no?
-Tom
-- Thomas Nichols -------------------- Department of Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
|