> I'm running a functional analysis of some data from brain-lesioned (stroke) patients and am posting to confirm the following steps that I should take in the first level analysis (in addition to those required for analyses of healthy participants). These are mostly gleaned from the forums, the wiki, and direct advice from MJ.
> 1) I should use an inverse mask of the lesion (registered to functional space) and input it in the post-stats tab under "pre-threshold masking" so that signal from lesioned areas will be excluded before thresholding.
That sounds reasonable.
> 2) I should include a timeseries from the lesioned area (registered to functional space) as a nuisance covariate by making a timeseries of the lesioned area as follows:
> fslmeants -i filtered_func_data.nii.gz -o my_meants.txt -m lesion_reg2_examplefunc.nii.gz.
> Then inputting it into the "add additional confound EVs" option of the stats tab.
I have not seen this approach before, but I guess it would be worth a shot. The assumption behind it would be that there is some common temporal behaviour of the voxels within the lesion and that by including that as a regressor one would be able to "mask out" voxels that one missed with the manually defined mask.
What you might want to be careful with is if within your mask there are viable tissue that actually participates in whatever task your subject is doing. You might then potentially regress out parts of your interesting signal from the rest of the brain.
> 3) To ensure the most accurate registration, I should FNIRT the structural image using an inverse lesion mask (so that the lesion doesn't warp the registration) as follows:
> fnirt --ref=MNI152_T1_2mm.nii --in=T1_brain.nii.gz --inmask=inverse_lesion_mask.nii.gz --iout=fnirted_T1_brain.nii.gz
> Then I should replace the highres2standard.mat file and the highres2standard_warp.nii.gz in the .feat/reg directory with the new ones from the masked fnirt output, and use updatefeatreg to incorporate the masked registration into the first-level analysis.
You should never run fnirt without a config-file so that call should be augmented by --config=T1_2_MNI152_2mm.cnf. You also need to pass in the affine transform from flirt so you need --aff="name_of_your_affine".mat.
> Is this appropriate? Is there anything that I've missed?
No, as far as I can see that is ok. Remember to mask out the lesion with "some margin", i.e. draw you lesion mask so that it extends a couple of voxels outside what you deem as lesion.