Dear Philippe,
> I am puzzled with the results of a conjunction analysis in SPM99. With
> the conjunction between two contrasts (A-B and C-D), I expect that the
> statistical map will show up only those regions commonly activated in
> these contrasts. However, the resulting statistical map show ALL the
> regions activated in the two contrasts, plus a p-corrected significant
> voxel which was not present in any of the initial contrasts! This is
> the case even when using an inclusive mask on each of these contrasts,
> with the significance level set up to 0.001 (uncorrected). The same
> conjunction in SPM96 (also using a mask) gave the expected result,
> i.e., a more restricted map where the activated voxels were already
> significantly activated in each of these contrasts. Can anybody
> explain where lies the difference between SPM96 and 99 in these
> results? Did I miss something ?
Conjunction analyses are implemented differently in SPM99 and SPM96.
In SPM99 you actually get the voxels that survive the specified
uncorrected p value threshold in each orthogonal contrast (if the
specified contrasts are not orthogonal they are rendered so
automatically after you enter an orthogonalization order). This
simpler and new approach was motivated by recent developments in
Gaussian field theory by Keith. These allow corrected p values to be
assigned to the intersection of the component SPMs (or equvalently the
SPM of the minimum T values over components). SPM96 did not have
access to these results and computed conjunctions as the sum of the
effects, specified by the contrasts, and removed voxels in which there
were significant differences among them. The new approach is much more
powerful in terms of corrected inference; in the sense that you are
constraining the search of one component SPM with the other[s]. Note
that with large numbers of orthogonal contrasts (e.g. 6) you only need
a very low threshold to get very significant results (e.g. 0.05 to the
power of 6 is 0.000000015625 and this is the chance uncorrected
probability of a conjunction).
With best wishes - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|