Compute MRIQC's metrics

[oesteban]

A prequel is nipreps/mriqc#440

[billbrod]

Has there been any progress on this, such that we can have fmriprep compute MRIQC's metrics or have MRIQC use the output of fmriprep when computing its metrics? It seems like MRIQC does a lot of the pre-processing steps that fmriprep does and so having them do them independently is wasteful. I couldn't find anything on the current documentation, but I might have missed it.

[effigies]

There has not. We're happy to help someone who wants to take this on, but I'm not sure that we're likely to get to it ourselves in the short term.

[billbrod]

Okay, we might try and look into it. Is there anything you can point me to in order to get started? Otherwise, my plan would be to just look through the steps of MRIQC and try to match them up to outputs / intermediates of fmriprep -- does that make sense?

[effigies]

Hi, yes. Sorry about the delay. I think the easiest places to look will be the main outputnode of the BOLD workflow:

fmriprep/fmriprep/workflows/bold/base.py

Lines 250 to 257 in c707938

	outputnode = pe.Node(niu.IdentityInterface(
	fields=['bold_t1', 'bold_t1_ref', 'bold2anat_xfm', 'anat2bold_xfm',
	'bold_mask_t1', 'bold_aseg_t1', 'bold_aparc_t1',
	'bold_std', 'bold_std_ref', 'bold_mask_std', 'bold_aseg_std', 'bold_aparc_std',
	'bold_native', 'bold_cifti', 'cifti_variant', 'cifti_metadata', 'cifti_density',
	'surfaces', 'confounds', 'aroma_noise_ics', 'melodic_mix', 'nonaggr_denoised_file',
	'confounds_metadata']),
	name='outputnode')

These should mostly be hooked up to derivatives:

fmriprep/fmriprep/workflows/bold/base.py

Lines 279 to 312 in c707938

	func_derivatives_wf = init_func_derivatives_wf(
	bids_root=layout.root,
	cifti_output=config.workflow.cifti_output,
	freesurfer=freesurfer,
	metadata=metadata,
	output_dir=fmriprep_dir,
	spaces=spaces,
	use_aroma=config.workflow.use_aroma,
	)
	workflow.connect([
	(outputnode, func_derivatives_wf, [
	('bold_t1', 'inputnode.bold_t1'),
	('bold_t1_ref', 'inputnode.bold_t1_ref'),
	('bold2anat_xfm', 'inputnode.bold2anat_xfm'),
	('anat2bold_xfm', 'inputnode.anat2bold_xfm'),
	('bold_aseg_t1', 'inputnode.bold_aseg_t1'),
	('bold_aparc_t1', 'inputnode.bold_aparc_t1'),
	('bold_mask_t1', 'inputnode.bold_mask_t1'),
	('bold_native', 'inputnode.bold_native'),
	('confounds', 'inputnode.confounds'),
	('surfaces', 'inputnode.surf_files'),
	('aroma_noise_ics', 'inputnode.aroma_noise_ics'),
	('melodic_mix', 'inputnode.melodic_mix'),
	('nonaggr_denoised_file', 'inputnode.nonaggr_denoised_file'),
	('bold_cifti', 'inputnode.bold_cifti'),
	('cifti_variant', 'inputnode.cifti_variant'),
	('cifti_metadata', 'inputnode.cifti_metadata'),
	('cifti_density', 'inputnode.cifti_density'),
	('confounds_metadata', 'inputnode.confounds_metadata'),
	('acompcor_masks', 'inputnode.acompcor_masks'),
	('tcompcor_mask', 'inputnode.tcompcor_mask'),
	]),
	])

So if you can find the derivative you want to generate summary statistics for, then you can hook it up from the outputnode. We'll probably want to create a new fmriprep/workflows/quality_control.py and init_func_qc_wf() function that takes inputs and generates output metrics.

Hopefully that can get you started, and then feel free to get in touch with any questions at all.