Dear Bernadette,
On Mon, 9 Oct 2017 09:37:05 +0200, Bernadette Rusch <[log in to unmask]> wrote:
>Dear all
>
>i am using the TFCE toolbox for the first time and was running into two
>questions:
>
>1) I usually use aal labeling for the clusters. However, when using the aal
>toolbox through the GUI I have no option to select TFCE cluster corrected
>outputs for the labeling. How do others do this or is there a better
>labeling option for the TFCE results?
I use the "Transform spmT maps" function in CAT12 for this purpose. There is an option to output the labels of predefined atlases that should be installed in SPM. Although this tool is expecting a filename such as spmT_00XX you can reset the file descriptor to also show other filenames. You should choose "Apply thresholds without conversion" and should use an already thresholded image to use this functionality. However, you should keep in mind the different and deviating atlas spaces in SPM and CAT12 (the latter is using DARTEL normalization and Dartel transformed atlases).
>
>2) I am using SPM12 and the TFCE toolbox. First I run a multiple regression
>analysis that also includes gender as an interaction term (calculated by
>multiplying the mean centered behavioral variable with the dichotomic one
>and including the new regressor in the multiple regression analysis), which
>results in a significant interaction effect (FWE whole brain corrected or
>TFCE FWE corrected).
>In order to follow-up on the direction of this effect, I run a multiple
>regression in each gender individually, masked for the clusters that were
>significant in the interaction across all. When I do this, I obtain whole
>brain FWE (p<0.05) corrected significant findings in one group. However, if
>i run the same contrast through the TFCE toolbox, no results remain. Is
>this even possible that a prior whole brain FWE corrected finding is not
>significant with the permutation based approach or is anything else in my
>methodology wrong?
Maybe, the clusters are too small to get sign. with TFCE if you limit to a ROI which was derived from a former analysis. I would check whether you get something sign. without restricting to the ROI...
Furthermore, you should check the the interaction indeed shows the main effect and should mask your interaction with this effect (or you check the plots to get sure that the interaction is in the right direction).
Best,
Christian
>
>Thanks a lot for any input!
>Best
>Bernadette
>
|