Hello,
I’m trying to decide about interesting IC's in Melodic using statistical criteria,
both in tensor and concatenation.
I’ve been founding some posts where you advice the use of fsl_glm command
with melodic_mix. However, I realize that tensor Melodic performs the test
with the rank-1 column of the t??.txt files, and these columns are different
from melodic_mix. I suppose that Melodic runs something like that:
fsl_glm -i <file with rank-1 column from t??.txt> -d design.mat -c design.con -
o <output> --out_z=<z_table> --out_p=<pvalue_table>
and outputs the results only in the web report page (which makes impossible
to threshold with some statistical criteria). The design.mat and design.con
were made with Glm gui with the EV’s from all participants concatenated in the
same order they entered Melodic. If there are few IC’s, it is not so
problematic, but if we have 200 or more, this is hard work.
In this case the solution is to use melodic_mix in the test:
fsl_glm -i melodic_mix -d design.mat -c design.con -o <output> --
out_z=<z_table> --out_p=<pvalue_table>
but the z and p-values will came different from the previous ones. Shouldn’t
they be the same? Which is the more correct to use?
On the other side, the t??.txt files made with concatenation Melodic do not
have the rank-1 column and in this case the second fsl_glm is the only
available. Is it correct to use that command in such situation?
With z and p-value matrices it is easy to import them into a spreadsheet and
establish some criteria over the contrasts to support the option for one IC and
not the other…
Kind regards,
José Paulo Santos
|