Automatically infer cohort only if infant output space is requested without cohort info

[tsalo]

A short summary of what you would like to see in NiBabies.

I think that the current approach in init_workflow_spaces will add MNIInfant with the appropriate cohort if no specific output spaces were requested by the user. I was thinking we could make this explicit in the output space specification, with something like MNIInfant (no cohort) or MNIInfant:cohort-auto (reserved value to infer the cohort).

That way you could have stuff like --output-spaces MNIInfant:cohort-1:cohort-auto MNI152NLin6Asym and it would still generate MNIInfant outputs with the age-appropriate cohort.

Do you have any interest in helping implement the feature?

Yes!

Add any additional information or context about the request here.

This has implications for other infant preprocessing pipelines, like QSIPrep (and maybe eventually ASLPrep).

[mgxd]

IMO, the output spaces syntax is already quite complicated, so I'm a bit hesitant to add another special value (auto). I like using MNIInfant, though some some changes to the spaces api will be needed for this.

Though, I think cases where multiple cohorts are requested should refer to the age ranges (and we should probably make that displayed somewhere rendered nicely) and explicitly add the cohorts of interest - the autodetection implemented was only meant to lower the barrier to entry.

And this brings to light another problem - I don't know if we have a consistent way to specify the internal common space when two or more MNIInfant cohorts are requested. I believe this should just be the first MNIInfant template provided within --output-spaces, but need to verify this is actually the case

[tsalo]

IMO, the output spaces syntax is already quite complicated, so I'm a bit hesitant to add another special value (auto).

For what it's worth, QSIPrep and QSIRecon need the ability to provide special values to output spaces. In their case, they need native-min and native-max options for resolution. @effigies proposed making it possible for applications to register spaces in nipreps/niworkflows#889, so additional special values would be managed on the BIDS App's side, not niworkflows'.

EDIT: The one issue with not using a special value is that it's a little harder to parse the output spaces. For example, if you want MNIInfant:cohort-1, MNIInfant:cohort-2 (this one's based on the age), and MNI152NLin6Asym outputs, --output-spaces MNIInfant:cohort-1:cohort-auto MNI152NLin6Asym is clearer than --output-spaces MNIInfant MNIInfant:cohort-1 MNI152NLin6Asym

I don't know if we have a consistent way to specify the internal common space when two or more MNIInfant cohorts are requested.

What is the "internal common space" in this scenario? The space used for figures like carpet plots?

[mgxd]

Dropped the ball on responding here, my bad

--output-spaces MNIInfant:cohort-1:cohort-auto MNI152NLin6Asym is clearer than --output-spaces MNIInfant MNIInfant:cohort-1 MNI152NLin6Asym

"MNIInfant:cohort-auto" will only be added if no MNIInfant output spaces are present - if a user is explicitly requesting 1 (or more), I assumed they looked up the bucket information (though come to think of it, we should include a nice table in the documentation!)

What is the "internal common space" in this scenario? The space used for figures like carpet plots?

The space for multistep registrations, basically native -> MNIInfant (antsRegistration) and MNIInfant -> MNI152NLin6Asym (and soon MNI152NLin2009cAsym) (precomputed), which are then combined into a single xfm.

[tsalo]

"MNIInfant:cohort-auto" will only be added if no MNIInfant output spaces are present - if a user is explicitly requesting 1 (or more), I assumed they looked up the bucket information (though come to think of it, we should include a nice table in the documentation!)

That makes sense. I can imagine scenarios where people want both, but that's probably an edge case more than anything. Supporting just MNIInfant (not cohort-auto) seems pretty reasonable.

The space for multistep registrations

Ah, so "internal common space" == "intermediate space" in the code.