Compatibility issue between Nipype preprocessing pipeline and FSL FEAT (1st level stats): Brainmask empty, data overly smooth

Dear all,

I am currently running a preprocessing pipeline up to 1st level FEATs, following Nipype’s FSL workflow (Neuroimaging in Python - Pipelines and Interfaces — nipy pipeline and interfaces package) and Michael Notter’s “hands on” session (handson_preprocessing).

My preprocessing pipeline is this:
i. brain extraction (struct)
ii. motion correction (func)
iii. robust intensity thresholding (excluding the highest and lowest 2% of intensity values) (func)
iv. smoothing (5mm Kernel, custom intensity and brightness thresholds) (func)
v. coregistration using FSL’s Boundary Based Registration (struct & func)
vi. global intensity normalisation (ensures the same median intensity value across all participants) (func)
vii. highpass filtering (func)
viii. motion/outlier detection using “art”

As far as I can see, everything turns out fine, yet if I run it from my resulting FEAT fsf file via the feat command line tool, the log file returns two unexpected statements:

Preprocessing:Stage 2

/media/storage/data/fsl/bin/fslmaths prefiltered_func_data filtered_func_data

/media/storage/data/fsl/bin/fslmaths filtered_func_data -Tmin -bin mask -odt char

/media/storage/data/fsl/bin/fslstats filtered_func_data -k mask -R | awk '{ print  }' -
calculateExtrema: mask is empty
calculateExtrema: mask is empty

/media/storage/data/fsl/bin/fslmaths filtered_func_data -Tmean mean_func

/bin/rm -rf prefiltered_func_data*


/media/storage/data/fsl/bin/smoothest -d 238   -m mask -r stats/res4d > stats/smoothness
WARNING: Extreme smoothness detected in X - possibly biased global estimate.
WARNING: Extreme smoothness detected in Y - possibly biased global estimate.
WARNING: Extreme smoothness detected in Z - possibly biased global estimate.

I assume the smoothness estimate would be a natural consequence given an empty mask, but FSL should definitely be able to find brains in the images.

I have written to Jeanette Mumford about this, who helpfully suggested that FSL cannot find the brain, if the data have already been mean centered.

However, maybe because my brain is fogged up by now^^, I am unable to see a step in my preprocessing that would cause mean centering. The obvious one, global intensity normalisation, is a recommended and necessary step in preprocessing as stated in the official FSL course under section 3, video 12 (FSL Course Material).
It is also, by default, executed in FSL.

I have checked the settings for my remaining preprocessing and none specify a setting or case in which mean centering would occur.

Does anyone have an idea why my brain masks should turn out empty?
Is it mean centering or possibly something else?

I would have loved to append the full log files, fsf designs and so on, but, according to Neurostars, I cannot append stuff as a “new user” …

Any help is greatly appreciated!

Hope you are all well and had a nice Easter time with your loved ones!
Stay safe and healthy!

Kind regards,