Running only functionals

[oesteban]

A good option to have would be running FMRIPREP without T1w images under two possible circumstances:

• The dataset does not have T1w images
• The user set a flag --dismiss-t1w to force dismissing the anatomical processing.

(comes from this discussion)

[madeleinesnyder]

Which script would be the best place to start with this? I tried changing run.py and just take away the raise exception when "No T1w images found", but then the workflow messed up. Is there a good junction to insert a "skip this part"?

[oesteban]

I think your target here would be writing an equivalent to init_single_subject_wf:

https://github.com/poldracklab/fmriprep/blob/a6fe5500ad2e49fb57dd6a4c98155be4a41570bd/fmriprep/workflows/base.py#L210-L462

where no anatomical workflows are instantiated and run. Without meaning to discourage you, I must note that this will have a steep learning curve :P. We'll try to help as much as we can, but we can't take this as a priority right now. I can see that in one month this would probably change.

[effigies]

In run.py, there's an --anat-only option, that you can imitate:

https://github.com/poldracklab/fmriprep/blob/a6fe5500ad2e49fb57dd6a4c98155be4a41570bd/fmriprep/cli/run.py#L87-L88

You could do --func-only or --bold-only or --dismiss-t1w (as @oesteban suggests). Any of these seems fine to me.

That will produce an opts.func_only (or bold_only or dismiss_t1w) that will need to be passed to init_fmriprep_wf:

https://github.com/poldracklab/fmriprep/blob/a6fe5500ad2e49fb57dd6a4c98155be4a41570bd/fmriprep/cli/run.py#L298-L327

This option will need to be added to the workflows in fmriprep/workflows/base.py, fmriprep/workflows/bold.py, and fmriprep/workflows/confounds.py.

The effects of --anat-only will be easy to mimic in base.py, and I think confounds.py should be pretty easy to pick out the confounds that require anatomical inputs and disable them.

bold.py is going to be the complicated one, as a bunch of subworkflows and connections should only be created if anatomicals are going to be used. You may want to look at how if freesurfer blocks work, as there are workflows that are based on whether or not the user wants to use FreeSurfer.

[madeleinesnyder]

Cool. I'm trying to puzzle out the bold_reg_wf because i think this is the place I want to say: "if dismiss_t1w, then do the bold_reg_wf with the mni brain and not the t1w because it doesn't exist. My confusion comes from where 't1_brain' is inputted to this workflow. Is there a master function coordinating the 't1_brain', 'bold_mask', etc things inside the workflow.connect function?

…

On Wed, Oct 11, 2017 at 12:50 PM, Chris Markiewicz ***@***.*** > wrote: In run.py, there's an --anat-only option, that you can imitate: https://github.com/poldracklab/fmriprep/blob/ a6fe550/fmriprep/cli/run.py#L87-L88 You could do --func-only or --bold-only or --dismiss-t1w (as @oesteban <https://github.com/oesteban> suggests). Any of these seems fine to me. That will produce an opts.func_only (or bold_only or dismiss_t1w) that will need to be passed to init_fmriprep_wf: https://github.com/poldracklab/fmriprep/blob/ a6fe550/fmriprep/cli/run.py#L298-L327 This option will need to be added to the workflows in fmriprep/workflows/base.py, fmriprep/workflows/bold.py, and fmriprep/workflows/confounds.py. The effects of --anat-only will be easy to mimic in base.py, and I think confounds.py should be pretty easy to pick out the confounds that require anatomical inputs and disable them. bold.py is going to be the complicated one, as a bunch of subworkflows and connections should only be created if anatomicals are going to be used. You may want to look at how if freesurfer blocks work, as there are workflows that are based on whether or not the user wants to use FreeSurfer. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#755 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJctLdNBWQz-ZC3AFWpzyGhTW3MyLklYks5srRv2gaJpZM4P15YA> .

[madeleinesnyder]

Or rather-- how are the input node fields (such as inputnode.t1_brain) populated? On Thu, Oct 12, 2017 at 11:42 AM, Madeleine Snyder < madeleine.snyder.48@gmail.com> wrote:

…

Cool. I'm trying to puzzle out the bold_reg_wf because i think this is the place I want to say: "if dismiss_t1w, then do the bold_reg_wf with the mni brain and not the t1w because it doesn't exist. My confusion comes from where 't1_brain' is inputted to this workflow. Is there a master function coordinating the 't1_brain', 'bold_mask', etc things inside the workflow.connect function? On Wed, Oct 11, 2017 at 12:50 PM, Chris Markiewicz < ***@***.***> wrote: > In run.py, there's an --anat-only option, that you can imitate: > > https://github.com/poldracklab/fmriprep/blob/a6fe5500ad2e49f > b57dd6a4c98155be4a41570bd/fmriprep/cli/run.py#L87-L88 > > You could do --func-only or --bold-only or --dismiss-t1w (as @oesteban > <https://github.com/oesteban> suggests). Any of these seems fine to me. > > That will produce an opts.func_only (or bold_only or dismiss_t1w) that > will need to be passed to init_fmriprep_wf: > > https://github.com/poldracklab/fmriprep/blob/a6fe5500ad2e49f > b57dd6a4c98155be4a41570bd/fmriprep/cli/run.py#L298-L327 > > This option will need to be added to the workflows in > fmriprep/workflows/base.py, fmriprep/workflows/bold.py, and > fmriprep/workflows/confounds.py. > > The effects of --anat-only will be easy to mimic in base.py, and I think > confounds.py should be pretty easy to pick out the confounds that > require anatomical inputs and disable them. > > bold.py is going to be the complicated one, as a bunch of subworkflows > and connections should only be created if anatomicals are going to be used. > You may want to look at how if freesurfer blocks work, as there are > workflows that are based on whether or not the user wants to use FreeSurfer. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#755 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AJctLdNBWQz-ZC3AFWpzyGhTW3MyLklYks5srRv2gaJpZM4P15YA> > . >

[oesteban]

They are set in by connecting this node:

https://github.com/poldracklab/fmriprep/blob/a6fe5500ad2e49fb57dd6a4c98155be4a41570bd/fmriprep/workflows/base.py#L363-L364

to those fields in the top-level workflow, eg.:

https://github.com/poldracklab/fmriprep/blob/a6fe5500ad2e49fb57dd6a4c98155be4a41570bd/fmriprep/workflows/base.py#L407-L408

sets the anatomical inputs.

[effigies]

To narrowly answer your question, t1_brain is set in bold_reg_wf in this line:

https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L353

I will say that registering BOLD to MNI is probably going to be the most difficult thing to attempt in our current setup. bold_reg_wf attempts to find affine transforms using FLIRT or mri_coreg/bbregister, which is then concatenated with a T1-MNI transform found by ANTs (in bold_mni_trans_wf).

If you want to estimate a nonlinear transform directly from BOLD to a template T2w or EPI image, that would probably be best handled by a new sub-workflow that bypasses both bold_reg_wf and bold_mni_trans_wf entirely, rather than repurposing these.

[madeleinesnyder]

Ok, my intuition is to insert: if dismiss_t1w: [new workflow code here] return workflow just before line 346 in bold.py is this on the right track?

…

On Thu, Oct 12, 2017 at 11:51 AM, Oscar Esteban ***@***.***> wrote: They are set in by connecting this node: https://github.com/poldracklab/fmriprep/blob/ a6fe550/fmriprep/workflows/ base.py#L363-L364 to those fields in the top-level workflow, eg.: https://github.com/poldracklab/fmriprep/blob/ a6fe550/fmriprep/workflows/ base.py#L407-L408 sets the anatomical inputs. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#755 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJctLe6hcm7-5ZZkNosV0XqzdK9otguyks5srl_PgaJpZM4P15YA> .

[oesteban]

Yes, that's the right direction. The problem, as Chris mentioned, resides on the fact that currently, BOLD workflows depend on anatomical inputs.

Since the anatomical workflows do not depend on BOLD workflows, it was relatively easy to use the approach you mentioned.

So the hard problem here is where you wrote: "[new workflow code here]" because you'll indeed need a new workflow (without dependencies on anatomical) here.

[madeleinesnyder]

Ok, Really the problem I want to solve is that if there are no T1w images available for a certain scan, the whole thing won't run at all. I wanted to create a functional only workflow to solve this problem.

…

On Thu, Oct 12, 2017 at 12:11 PM, Oscar Esteban ***@***.***> wrote: Yes, that's the right direction. The problem, as Chris mentioned, resides on the fact that currently, BOLD workflows depend on anatomical inputs. Since the anatomical workflows do not depend on BOLD workflows, it was relatively easy to use the approach you mentioned. So the hard problem here is where you wrote: "[new workflow code here]" because you'll indeed need a new workflow (without dependencies on anatomical) here. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#755 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJctLX5iggrHZr1Xxup6ZB_w8UMOzcp9ks5srmR5gaJpZM4P15YA> .

[effigies]

I'll try to pare this down to the minimum things you need to do just to get it running, and we can work from there to build out the remaining needed functionality, such as registration to MNI. Assuming you have a variable dismiss_t1w.

Split this chunk:

https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L346-L390

Into two bits, to be schematic as possible:

workflow.connect([
    # Only connect workflows that do not require T1w processing
    ])
if not dismiss_t1w:
    workflow.connect([
        # Connect up the rest
        ])

These lines take the following form:

(from_node, to_node, [('from_name', 'to_name')]),

Nodes that do not require T1w preprocessing are:

• inputnode, outputnode
• bold_reference_wf
• bold_stc_wf
• bold_hmc_wf
• func_reports_wf
• summary

For the moment, I would entirely disable fieldmaps by commenting out:

https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L413-L480

And replace with:

if True:

Comment these lines out, as well:

https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L487-L489
https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L492-L579

Then pacify summary:

https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L280-L286

by adding the following lines:

if dismiss_t1w:
    summary.inputs.fallback = False
    summary.inputs.confounds_list = []

Finally, disable some reports by commenting out these lines:

https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L1605-L1612

I think that should technically run. You will get some minimal reports, but no derivatives. If you want BOLD series in their original BOLD space, that will need a new workflow, as will an MNI registration workflow.

Once you have this, start a PR, and we can comment directly on that, which should make things easier.

[madeleinesnyder]

Wow this a lot to go on thank you so much! I'll start cracking

…

On Thu, Oct 12, 2017 at 12:35 PM, Chris Markiewicz ***@***.*** > wrote: I'll try to pare this down to the minimum things you need to do just to get it running, and we can work from there to build out the remaining needed functionality, such as registration to MNI. Assuming you have a variable dismiss_t1w. Split this chunk: https://github.com/poldracklab/fmriprep/blob/ 0e2539e/fmriprep/workflows/ bold.py#L346-L390 Into two bits, to be schematic as possible: workflow.connect([ # Only connect workflows that do not require T1w processing ])if not dismiss_t1w: workflow.connect([ # Connect up the rest ]) These lines take the following form: (from_node, to_node, [('from_name', 'to_name')]), Nodes that do not require T1w preprocessing are: - inputnode, outputnode - bold_reference_wf - bold_stc_wf - bold_hmc_wf - func_reports_wf - summary For the moment, I would entirely disable fieldmaps by commenting out: https://github.com/poldracklab/fmriprep/blob/ 0e2539e/fmriprep/workflows/ bold.py#L413-L480 And replace with: if True: Comment these lines out, as well: https://github.com/poldracklab/fmriprep/blob/ 0e2539e/fmriprep/workflows/ bold.py#L487-L489 https://github.com/poldracklab/fmriprep/blob/ 0e2539e/fmriprep/workflows/ bold.py#L492-L579 Then pacify summary: https://github.com/poldracklab/fmriprep/blob/ 0e2539e/fmriprep/workflows/ bold.py#L280-L286 by adding the following lines: if dismiss_t1w: summary.inputs.fallback = False summary.inputs.confounds_list = [] Finally, disable some reports by commenting out these lines: https://github.com/poldracklab/fmriprep/blob/ 0e2539e/fmriprep/workflows/ bold.py#L1605-L1612 I think that should technically run. You will get some minimal reports, but no derivatives. If you want BOLD series in their original BOLD space, that will need a new workflow, as will an MNI registration workflow. Once you have this, start a PR, and we can comment directly on that, which should make things easier. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#755 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AJctLfQWWMy2UYv9Fh0KJkkeSqhB_OqTks5srmoMgaJpZM4P15YA> .

[madeleinesnyder]

I thought of another workaround that might be easier: Would it be ok to use a dummy T1 dataset as a placeholder to get the pipeline to run, and have the --output-space be template so that the dummy T1 isn't used for coregistration? Or would the dummy be automatically used in other parts of the pipeline even if the output space is "template"? On Thu, Oct 12, 2017 at 12:41 PM, Madeleine Snyder < madeleine.snyder.48@gmail.com> wrote:

…

Wow this a lot to go on thank you so much! I'll start cracking On Thu, Oct 12, 2017 at 12:35 PM, Chris Markiewicz < ***@***.***> wrote: > I'll try to pare this down to the minimum things you need to do just to > get it running, and we can work from there to build out the remaining > needed functionality, such as registration to MNI. Assuming you have a > variable dismiss_t1w. > > Split this chunk: > > https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6 > e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L346-L390 > > Into two bits, to be schematic as possible: > > workflow.connect([ > # Only connect workflows that do not require T1w processing > ])if not dismiss_t1w: > workflow.connect([ > # Connect up the rest > ]) > > These lines take the following form: > > (from_node, to_node, [('from_name', 'to_name')]), > > Nodes that do not require T1w preprocessing are: > > - inputnode, outputnode > - bold_reference_wf > - bold_stc_wf > - bold_hmc_wf > - func_reports_wf > - summary > > For the moment, I would entirely disable fieldmaps by commenting out: > > https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6 > e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L413-L480 > > And replace with: > > if True: > > Comment these lines out, as well: > > https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6 > e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L487-L489 > https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6 > e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L492-L579 > > Then pacify summary: > > https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6 > e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L280-L286 > > by adding the following lines: > > if dismiss_t1w: > summary.inputs.fallback = False > summary.inputs.confounds_list = [] > > Finally, disable some reports by commenting out these lines: > > https://github.com/poldracklab/fmriprep/blob/0e2539e0cd472c6 > e225d73e0c62f91b798131544/fmriprep/workflows/bold.py#L1605-L1612 > > I think that should technically run. You will get some minimal reports, > but no derivatives. If you want BOLD series in their original BOLD space, > that will need a new workflow, as will an MNI registration workflow. > > Once you have this, start a PR, and we can comment directly on that, > which should make things easier. > > — > You are receiving this because you commented. > Reply to this email directly, view it on GitHub > <#755 (comment)>, > or mute the thread > <https://github.com/notifications/unsubscribe-auth/AJctLfQWWMy2UYv9Fh0KJkkeSqhB_OqTks5srmoMgaJpZM4P15YA> > . >