Hi,
for those running dual regression but then wanting to do the randomise comparisons as a separate stage of analysis, the advice below may be useful (thanks to Anderson Winkler):
question:
i have successfully run the dual reg with all my scans in one large group, specifying NOT to perform randomise (my intent only to generate the dr_stage2 output for each scan). my plan was then to run randomise separately for each of the 3 comparisons of interest, now prepping for the randomise,i realise i am uncertain how to prepare my input image file? reading the dual_reg and randomise wiki it is clear that it should be the dr_stage2 data, either the _ic or _subject-specific files. the difficulty i now realise i have, is that for example the dr_stage2_ic0000 file contains a volume for each of the original scans that i fed into the dual reg (40 in total), but that this includes both my paired scans (one before and after scan for each of the bipolar subjects AND one for each of the healthy controls). if i wish to feed this into the randomise, it wouldn't correspond to any of my 3 comparisons, i.e. 1st comparison would be before vs after scan from the bipolar subjects but musn't include the healthy controls. since the instructions for randomise specify a single 4D input image, do i need to fslsplit the dr_stage2_ic files, remove whichever vols don't correspond to the particular comparison i am busy with, fslmerge back into one file for _ic, and then fslmerge again all the dr_stage2_ic files into one larger file for randomise - problem with this, is that the internal arrangement of data within the input file no longer would correspond to the GLM?
answer:
Right, you need to use fslsplit to break these 4D files apart, then use fslmerge (-t) to merge those that matter for each of the 3 comparisons. For the comparisons HC vs BP1 and HC vs BP2, if the number of subjects is the same, and the files are organised in the same order, the design files for the GLM can probably be the same. For the comparison BP1 vs BP2 you need a paired t-test, so the GLM will look different anyway. The best thing to do is really just fslsplit then fslmerge (3x) then prepare the GLM files (3x) then run randomise (3x) FOR EACH OF THE ICs. You may want to write a short shell script to do this all automatically for you.
|