problem applying niftimasker

[LCD12345]

Hi

I am trying to learn about MVPA and nilearn and am encountering a problem. I have an event-related design so I followed the recommendation to create trial-specific regressors (used SPM 12), which gave me a beta image for each trial event.

I created a mask using SPM 12 neuromorphometric atlas in order to restrict the analysis to an ROI. When I come to use NiftiMasker to apply this mask to any of the beta images it provides this warning:

C:\Users\L\AppData\Roaming\Python\Python37\site-packages\nilearn\image\resampling.py:584: RuntimeWarning: NaNs or infinite values are present in the data passed to resample. This is a bad thing as they make resampling ill-defined and much slower.
  fill_value=fill_value)
C:\Users\L\AppData\Roaming\Python\Python37\site-packages\nilearn\image\resampling.py:591: RuntimeWarning: invalid value encountered in greater
  vmin = min(data.min(), 0)
C:\Users\L\AppData\Roaming\Python\Python37\site-packages\nilearn\image\resampling.py:592: RuntimeWarning: invalid value encountered in less
  vmax = max(data.max(), 0)
C:\Users\L\AppData\Roaming\Python\Python37\site-packages\nilearn\signal.py:61: UserWarning: Standardization of 3D signal has been requested but would lead to zero values. Skipping.
  warnings.warn('Standardization of 3D signal has been requested but '

I believe this has something to do with how SPM 12 uses NaN to represent the values outside of the brain, but I could be wrong. I tried nilearn.image.clean_img() but received an error because this expects a 4D array - my data are 3D betas since it was event related. In any case, I see that there is the argument ensure_finite when calling the apply_mask() which is by default True, so then I'm not sure why I receive this warning/error.

I guess I could read in the data using nibabel.load() and then manually set the NaN values to zeros but I want to apply the mask to all of my betas and the NiftiMasker expects a filename rather than an array. Maybe I could do this and re-save the beta file.

I am unable to attach a couple of my beta.nii but happy to email if it helps. Here is a link to a notebook that has a minimal example with outputs to see - hope it is helpful.

[bthirion]

Normally, if you run nilearn.image.clean_img() on the list of all images,this is equivalent to running it on a 4D image. This should do the trick. Can you confirm ?
Also, we should have this kind of discussion on the Neurostars site rather than the Nilearn issue tracker.
Best,

[LCD12345]

Also, we should have this kind of discussion on the Neurostars site rather than the Nilearn issue tracker.
Best,

Ok, going back to the thread.