[Neuroimaging] [fmriprep] Specifying different priors and templates

[Chris Gorgolewski]

Hi Dan!

It would be great to get your help! More comments below:

On Tue, Jan 10, 2017 at 1:07 PM, Dan Lurie <dan.lurie at berkeley.edu> wrote:

> Hey fmriprep team,
>
> I am working on a pull request to add support for using lesion masks
> during T1 registration, and wanted to check in with you guys before I got
> too deep. A couple of questions:
>
> 1) I’m assuming it’s best to create a separate new pipeline based on ds005
> instead of modifying ds005? I want to make sure i’m adding things in a way
> that is consistent with the overall architecture you’ve envisioned.
>
Yes this is the best way to go around it. In general FMRIPREP is looking at
what data is available for a given subject and then applies the appropriate
pipeline (this happens here
<https://github.com/poldracklab/fmriprep/blob/master/fmriprep/workflows/base.py#L56>).
So in your case you should see if there is a lesion mask available for a
given subject and chose your new workflow instead of ds005. The selection
of the workflow can be overridden on the commandline - os you need to add
the name of the new workflow here
<https://github.com/poldracklab/fmriprep/blob/master/fmriprep/run_workflow.py#L59>.
BTW we are considering changing workflow names
<https://github.com/poldracklab/fmriprep/issues/311> to make them more
intuitive.

The new workflow should reuse subworkflows of ds005 as much as it can. We
want to minimize the amount of replicated code.

2) Lesion masks are usually in subject native space, so using cost function
> masking in ANTs requires specifying the template as the moving image and
> then using the inverse transforms (see here: https://github.com/
> stnava/ANTs/issues/48). My plan was to just change the pipeline workflow
> inputs/outputs to accommodate this change (e.g. use the inverse transform
> when registering the EPI and TPMs). Is there anything tricky about how the
> pipeline is organized that I should keep in mind when doing this?
>
I see. You just need to connect the output transforms to the right
outputnode fields. You also minght need to calculat/generate inverse
transforms.


> 3) is run_workflow.py the place to add options/settings for things like
> specifying what templates to use during skull stripping and tissue
> segmentation? I looked through the BIDS-Apps docs to see if there was any
> common design pattern established, but didn’t see anything.
>
If you want to add an option to use different templates for
BrainsExtraction doing it via a commandline parameter in run_workflow.py is
the best option. Is the template we are using now (OASIS) not working for
your data? Maybe we can figure out how to choose which template to use
based on some data properties?

Thanks in advance, and apologies if I’m missing anything obvious. Happy to
> chat over Skype/Hangouts if that’s easier than email.
>
Happy to have a chat if you want to. We should get some documentation
for develeopers
soon <https://github.com/poldracklab/fmriprep/issues/309>.

BTW one thing I'm interested in is how did you organize your input data.
The lesion masks are hand drawn yes? We should work on some standardized
way of saving such data in BIDS that FMRIPREP can work with.

Best,
Chris

> Dan
>
> --
> Dan Lurie
> Graduate Student
> Department of Psychology
> University of California, Berkeley
> http://despolab.berkeley.edu/lurie
>
> _______________________________________________
> Neuroimaging mailing list
> Neuroimaging at python.org
> https://mail.python.org/mailman/listinfo/neuroimaging
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/neuroimaging/attachments/20170110/56ec6549/attachment.html>