[6VV2] AFNI through NIpype (AFNI - raw)

[JLefortBesnard]

Softwares

AFNI

Input data

raw data

Additional context

I attempted to replicate the pipeline 6VV2 using three different methods for the first level (preprocessing + subject-wise regression) and the same three methods for the second level (group-wise regression)
1.Recreate the AFNI code using Nipype AFNI interface
2.Execute the afni_proc.py file through Nipype (not using the AFNI interface)
3.Execute the afni_proc.py file independently.

Issues with method 1

Some Afni features are not implemented in the Afni interface of Nipype

• Nipype deconvolve function doesn't have the "stim_times_AM2" options of the afni deconvolve function

• When a temporary file should be transposed, nipype workflow does not allow for this. ex: AFNI: 3dTstat -sos -prefix - mean.errts.unit.1D\' > out.gcor.1D or set blurs = (`3dTstat -mean -prefix - blur.epits.1D'{0..$(2)}'\'` )

• Missing gen_epi_review.py

• Missing auto_warp.py

• 3dbucket in nipype requires list of tuple with 2 elements (for 6VV2 only one is needed: 3dbucket -DAFNI_NIFTI_VIEW=tlrc -prefix sub-001_T1w_ns awpy/sub-001_T1w_ns.nii*

Some Afni features are not implemented with the same options in the Afni interface of Nipype

• 3dmean as afni.Means() takes only 2 inputs (in_file_a and in_file_b) instead of more than 2

• 3dDot instead of 3ddot in nipype (written 3dDot when actually calling the 3ddot afni function thus returns an error)

• The nipype version of align_epi_anat.py does not have the -skullstrip attribute working (error returned: AttributeError: 'dict' object has no attribute 'skullstrip')

Issues with method 2

Using a computer with 16G Ram, the code systematically crashed (Ram saturation)

Suggested solutions

As I proceeded, within the script, include sections utilizing Nipype workflows, and other sections directly invoking Afni functions (e.g., "!3d mean").
Another option is to modify the nipype afni interface locally
A greater option is to contribute to the nipype repro directly

[bclenet]

Here are a few notes, for further work on these issues.

• 3dDeconvolve : args input of the nipype.interfaces.afni.model.Deconvolveinterface could allow to add the missing parameter stim_times_AM2
• 3dtstat : I need to better understand the issue
• gen_epi_review.py : this would require to develop a new nipype interface. Is this feature really needed in our usecase ?
• auto_warp.py : this would require to develop a new nipype interface.
• 3dBucket : not sure to understand this, but it looks like the required tuple is on the form (file: str, subbrick_selection: str), perhaps using '{0..$}' as subbrick_selection to select all ?
• 3dMean : args input of the nipype.interfaces.afni.preprocess.Meansinterface could allow to add the other files (to be tested though)
• 3dDot : workaround (no extra code needed) - modify the _cmd attribute of the interface, e.g.: dot._cmd = '3ddot'
• align_epi_anat.py : this would require to develop a new nipype interface.