I was running a lateralization analysis of task fmri data, so I needed to obtain R-L unflipped and flipped copes through registration to a symmetric template, to enter in higher-level feat, both at subject level (i.e second level) and , then, at group level (third level). The easy way was to run full analysis by starting from not pre-processed images separately, as this would obtain complete .feat directories with the desired copes and related /reg directories. plus the fact that the T1 corresponding to initial flipped 3D input has to be flipped too to obtain the correct transformation matrixes to the symmetric template that substitutes the standard MNI brain.
The issue I faced is that the two full 1st-level analyses, for unflipped and the same flipped 3d data (fslswapdim data -x y z flip_data) as input, gave slightly different results. I checked (fslstats -R -M) and I found differences already present at the filtered_func_data, The mcflirt parameters were slightly different (0.27 vs 0.28 for example). I can’t see why this happens, if it’s a bug or there are random factors in the functioning of some modules of the feat pipeline.
Could you please comment on that (unexpected?)difference, if it is a methodollogical issue and propose some resolution for using, possibly, Feat only (for adherence to the parametric analysis)?
For example, many authors flip copes, but those cannot be entered into a feat high-level. In that case there would be needed to make the simple fslmaths unflipped-minus-flipped subtraction and then use randomise for group comparisons. In that case, is any smoothing proposed too?
Thanks a lot!
Nikolas
Post-doc,Sapienza University, Rome, Italy
|