No, I loaded the input data files into Matlab and there appear to be no NaNs
in the input to the first level analysis. Nor were there any NaNs in the
cope1.nii.gz, cope2.nii.gz, etc. files generated by the first-level analysis.
I tried using fslmaths to get rid of the NaNs in the inputs to the
third-level analysis and after I did that, nothing passed the cluster
threshold at the third level (despite the fact we had a number of good-sized
areas of activation before getting rid of the NaNs).
This makes me concerned about whether and how this problem effects the
inferences we made from the original (NaN-containing) analysis.
We're using the most recent version of FSL 4, the one with the updated
masks. I don't seem to have had this problem in older versions.
|