24.0.0rc fails with `'MNIInfant not found in SpatialReferences` if ``cifti-output`` flag not used

[scott-huberty]

EDIT: Edited this message on Aug 6th to provide more background info.

Background

#360 by @mgxd Added this function:

nibabies/nibabies/workflows/base.py

Lines 953 to 964 in 38fb3fd

	def get_MNIInfant_key(spaces: SpatialReferences) -> str:
	"""Parse spaces and return matching MNIInfant space, including cohort."""
	key = None
	for space in spaces.references:
	# str formats as <reference.name>:<reference.spec>
	if 'MNIInfant' in str(space) and 'res-native' not in str(space):
	key = str(space)
	break
	if key is None:
	raise KeyError(f'MNIInfant not found in SpatialReferences: {spaces}')
	return key

Used here:

nibabies/nibabies/workflows/base.py

Lines 517 to 525 in 38fb3fd

	if 'MNIInfant' in [ref.space for ref in spaces.references]:
	select_MNIInfant_xfm = pe.Node(
	KeySelect(
	fields=['anat2std_xfm', 'std2anat_xfm'],
	key=get_MNIInfant_key(spaces),
	),
	name='select_MNIInfant_xfm',
	run_without_submitting=True,
	)

---

The Problem

If config.execution.output_spaces is None (which I believe is the default value), then Nibabies will add MNIInfant:cohort-1:res-native at this LOC:

nibabies/nibabies/workflows/base.py

Lines 829 to 832 in 38fb3fd

	if not spaces.references:
	# Ensure age specific template is added if nothing is present
	cohort = cohort_by_months('MNIInfant', age_months)
	spaces.add(('MNIInfant', {'res': 'native', 'cohort': cohort}))

Thenget_MNIInfant_key will throw an error on that exact value that Nibabies just added on Line 830.... I.e.. for release candidate 24.0.0rc, a lot of my files that ran without issue on 23.1.0 now always hit the error:

KeyError: 'MNIInfant not found in SpatialReferences: Spatial References: MNIInfant:cohort-1:res-native'

Full Stack Trace

Traceback (most recent call last):
  File "/opt/conda/envs/nibabies/bin/nibabies", line 8, in <module>
    sys.exit(main())
             ^^^^^^
  File "/opt/conda/envs/nibabies/lib/python3.11/site-packages/nibabies/cli/run.py", line 62, in main
    retval = build_workflow(config_file)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/nibabies/lib/python3.11/site-packages/nibabies/cli/workflow.py", line 76, in build_workflow
    retval['workflow'] = init_nibabies_wf(config.execution.unique_labels)
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/nibabies/lib/python3.11/site-packages/nibabies/workflows/base.py", line 147, in init_nibabies_wf
    single_subject_wf = init_single_subject_wf(
                        ^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/nibabies/lib/python3.11/site-packages/nibabies/workflows/base.py", line 521, in init_single_subject_wf
    key=get_MNIInfant_key(spaces),
        ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/nibabies/lib/python3.11/site-packages/nibabies/workflows/base.py", line 963, in get_MNIInfant_key
    raise KeyError(f'MNIInfant not found in SpatialReferences: {spaces}')
KeyError: 'MNIInfant not found in SpatialReferences: Spatial References: MNIInfant:cohort-1:res-native'

---

• This error isn't very helpful because 1) MNIInfant is in Spatial References, and 2) there's no further info on what should be changed.

• I checked locally, and on 23.1.0, the value of my Spatial Reference is the same: MNIInfant:cohort-1:res-native. The only difference on 24.0.0rc seems to be that this new function, get_MNIInfant_key forces an error on this value.

• If I modify the function to accept this value, then things seems to run okay again. But I'd like to understand what get_MNIInfant_key really expects before I submit a PR for what I presume to be a bug.

[mgxd]

IIRC, this was to deal with the MNIInfant -> MNI152NLin6Asym transformation required for CIFTI outputs, since we are using a saved transform from a 1x1x1 isotropic version of each. My guess is you aren't using the --cifti-output flag, hence running into that. Seems like we need to add a check to ensure this is not required if not requesting CIFTI outputs.

[scott-huberty]

IIRC, this was to deal with the MNIInfant -> MNI152NLin6Asym transformation required for CIFTI outputs, since we are using a saved transform from a 1x1x1 isotropic version of each. My guess is you aren't using the --cifti-output flag, hence running into that. Seems like we need to add a check to ensure this is not required if not requesting CIFTI outputs.

Yeah you are exactly right. I can confirm that I was not using the --cifti-output flag, and if I do use it, I don't hit the error.

[mgxd]

@scott-huberty do you want to submit a PR? otherwise i can address this

[scott-huberty]

@scott-huberty do you want to submit a PR? otherwise i can address this

Sure! I'll give it a shot. Feel free to assign me.