Finding appropriate fsl residuals file for 3dFWHMx

I am using 3dClustSim to conduct cluster correction for an analysis run using FSL’s FEAT. I need to estimate the smoothness of my data using 3dFWHMx.

So in Afni, when you complete first level analysis with, you get a residual file typically called errts+tlrc. This is the residual of the raw time series after you account for all your model parameters, and what is recommended to use as input to estimate the smoothness of the data using 3dFWHMx

In our lab we use fsl FEAT, which also outputs a residual file called res4d.nii

BUT there is an important difference in how first level analyses with multiple runs are typically analyzed in afni and fsl. In afni, you concatenate the timeseries, then run the analysis. In fsl, you run the analysis separately for each run, then average those analyses in a gfeat. You therefore have a res4d.nii file for each run (in this task 4 runs) and one for the gfeat model that averages the models for each run.

The res4d.nii files from each of the 4 run are residual timeseries of that run.
The res4d.nii files from the gfeat are the residuals of the model averaging the 4 runs, so they just contain 4 data points per voxel that sum to 0.

My question is: what should I use as input to 3dFWHMx?


Hi, David-

Well, you could always be the groundbreaker and lead the charge into land! I mean, is pretty fun:

Barring that, my guess is that you might run 3dFWHMx on each of the 4 resd.nii residual time series for each subj; certainly, within a subject, the smoothness estimates shouldn’t vary very much— you should check them and verify that. If they do vary a lot, then something weird is going on. You could then average the parameter estimates across runs for that subject.

We normally average each parameter across the group results, and those tend to be quite similar (for data acquired on the same scanner/in a similar way). This was recently discussed a little bit in a separate question, here:,163697,163723#msg-163723

Does that seem reasonable?


Yes. This is very helpful. Thank you!