Add option to alternative MNI templates for output spaces

[feilong]

From discussions of another issue (nipreps/fmriprep#337 (comment)), it would be nice to have alternatives to MNI152NLin2009cAsym as output spaces to increase backward compatibility with other softwares, datasets, and atlases. For example, currently widely used MNI152NLin6Asym.

I'll try to modify fmriprep/workflows/anatomical.py and add output_spaces as an input parameter to init_anat_preproc_wf, and change the spatial normalization code accordingly. Does that sound good? @chrisfilo

[effigies]

If I can make a suggestion, what might be nice would be to add a --template-space={MNI152NLin2009cAsym,MNI152NLin6Asym,<etc>} and change the MNI option in --output-space to template. This keeps --output-space options to a reasonably short list of manageable strings.

Unless we think people are potentially going to want their data in many template spaces?

[chrisgorgo]

@feilong this change would require modifications of the RobustMNINormalization Interface in niworkflows: https://github.com/poldracklab/niworkflows/blob/master/niworkflows/anat/mni.py#L35. The target atlas would also have to be included in the data downloader https://github.com/poldracklab/niworkflows/blob/master/niworkflows/data/getters.py#L19 (Although @oesteban might know if it is already included - if not if you can email it to me I will add it to our OSF repository).

@effigies Your proposal does reflect the fact that only one volumetric template would be allowed (which should be fine for 99% of users). I would however insist on keeping the full names of the template used in output file names (instead of "template") for provenance related reasons..

[effigies]

Yeah. Keeping the full name in filenames makes sense. Just from a UI perspective, not having to write out the full template unless changing from the default would be nice.

[effigies]

Actually, speaking of this, should creating structurals in MNI space depend on --output-space? Currently t1_2_mni and the mni_* nodes run no matter what.

[feilong]

@effigies I think so, and I guess it might be good to add template space names or abbreviations to those node names?

[oesteban]

@feilong do you refer to this atlas -> http://nist.mni.mcgill.ca/?p=858 ? It is not currently included in niworkflows. As @chrisfilo said, you'd need to add it.

[chrisgorgo]

MNI152NLin6Asym is an asymmetrical version of this atlas - we would have to grab it from SPM or FSL.

[chrisgorgo]

@feilong I'm trying to prioritize features - would you use/benefit from addition of MNI152NLin6Asym?

[feilong]

@chrisfilo Yes, I want my data resampled to both MNI152NLin6Asym and MNI152NLin2009cAsym if possible.

[effigies]

@chrisfilo is it worth updating --template to take multiple template specifiers?

[chrisgorgo]

Interesting. So you would like outouts in both MNI152NLin6Asym and MNI152NLin2009cAsym? Despite it would take double the space?

We could either iterate over T1w coregistration targets or always use MNI152NLin2009cAsym and in the final stage apply one more non-linear warp from MNI152NLin2009cAsym to MNI152NLin6Asym. Such warp should not be hard to estimate.

[feilong]

I need some compatibility at this moment, but eventually MNI152NLin2009cAsym would be the choice.
I guess coregister to the target template directly would be more more accurate, though it would take a little longer.

[dangom]

Hi guys, would you have any updates on this issue? I was checking out fmriprep and was a bit frustrated that the template is fixed. For resting-state studies, I'd like to use a dual regression approach against maps generated in MNI 6th gen 2mm. I don't know the consequences of doing this if my maps are in MNI 2009.

Besides that, ICA AROMA hyperplane parameters were developed for MNI 6th 2mm, as seen in the code here and here and in the Manual. Does this mean that, when the --usa-aroma flag is on, registration to MNI 6th is actually performed but then discarded? I reckon that the technique may also work in other templates, but it'd be nice to document that.

Finally, would there be a way of quickly hacking fmriprep into using MNI 6th 2mm by overwriting the MNI 2009 file? Any ideas if things would break?

(by the way, thanks a lot for this project)

[chrisgorgo]

Hi guys, would you have any updates on this issue? I was checking out fmriprep and was a bit frustrated that the template is fixed. For resting-state studies, I'd like to use a dual regression approach against maps generated in MNI 6th gen 2mm. I don't know the consequences of doing this if my maps are in MNI 2009.

Considering usual level of smoothness I would not expect a large difference, but we have not tested this explicitly. In principle we have nothing against adding an option to support other target templates (as you probably noticed the command line flag have been designed to accommodate that feature in the future). This could be something that you can help us to add by submitting a pull request.

Besides that, ICA AROMA hyperplane parameters were developed for MNI 6th 2mm, as seen in the code here and here and in the Manual. Does this mean that, when the --usa-aroma flag is on, registration to MNI 6th is actually performed but then discarded? I reckon that the technique may also work in other templates, but it'd be nice to document that.

No - as indicated by the file naming convention of the outputs, transformations to 2009c template are used in the AROMA pipeline. I would be surprised if using MNI 6th would make a significant difference in this context.

Finally, would there be a way of quickly hacking fmriprep into using MNI 6th 2mm by overwriting the MNI 2009 file? Any ideas if things would break?

It would break my heart :) We would much more appreciate if you could work towards implementing this as part of the core package extending the --template flag. We can provide guidance in the process.

(by the way, thanks a lot for this project)
Thank you for the kind words and the feedback. You might also read this thread on neurostars

• https://neurostars.org/t/bids-fmriprep-transforms-to-3-5-x-3-75-x-3-75-mni-space/1391
• https://neurostars.org/t/fixing-inconsistent-output-spaces-across-studies-in-fmriprep-1-0-6/1309

[jdkent]

It would break my heart :)

😆