OUTPUT SINK:
<sink_dir>/subject-//<source_entities>_.
ex:
pumi-results/sub-001/anat/sub-001_T1w_brain.nii.gz
pumi-results/sub-001/anat/sub-001_T1w_brain_mask.nii.gz
pumi-results/sub-002/func/sub-001_task-rest_bold_mc.nii.gz

suffix: always defined by the subworkflow (we might be able to re-define it)

QC SINK:
//<subworkflow_name>/<source_entities>_.

<source_entities>: must contain the subject id

ex:
qc/anat/bet/sub-001_T1w_brain.ppm
qc/anat/bet/sub-001_T1w_brain_mask.ppm
pumi-results/qc/func/mc/sub-001_task-rest_bold_mc_rot.png

For some methods, documentation still needs to be written.

Check if the container created from the Dockerfile really works.
To do this, run a pipeline in the container.

The plotting functions in connectivity.py (plot library) must be tested.
Perhaps they are not yet general enough.

The standard qc-pipeline regex substitution not only adjusts the path to the desired format, but also replaces the filename with 'sub-<subject_id>.png'.

It should be possible to sink other formats as well (i.e. the actual file extension should be catched and appended to the end)

Add exception(s) to the HD-Bet interface in case a file is passed that is not of type .nii.gz

Necessary for the integration of the RPN-signature

In the func_preproc library is Onevol.py which still needs to be ported to iterables.

port anat2mni_fsl from the old PUMI repo

If the comments are needed later, we still have them in the old PUMI repo.

The decorated function gets access to a PUMI.NestedWorkflow instance, passed in the function signature as wf.

PUMI workflows accept "default" strings in some cases, like:

wf.connect('inputspec',` 'in_file', despike, 'in_file')

Let's discuss if it makes sense to accept wf.inputspec or pass inputspec as a function parameter instead.

This applies to outputspec and sinker, too.

Allow something like:
head = data/standard/MNI152_T1_1mm.nii.gz; source=templateflow

parameter 'ref_vol' should be settable to 'mean'

• pass kwargs to most straightforward node constructor
• If there is more than one straightforward option: pass multiple dicts that follow the convention <software_tool>_kwargs

only after the iterable vs. mapnode discussion

take the BET node from the old PUMI repo and "port" it to the new PUMI.
Adjust it based on the (possible) decision of using iterables instead if MapNodes for subjects.

Necessary for the integration of the RPN-signature

Only do this after prior discussion

QC outputs needs to be sinked in a special structure to support easy quality inspection.

Necessary for the integration of the RPN-signature

change the name of globals.py to default.py

not needed anymore

The brain extractions do not look outstanding.
The vertical gradient probably needs to be adjusted.

fnirt_config-fallback missing in anat2mni_fsl

Only possible after the discussion about the style and naming convention

Adapt the code (especially for connecting the nodes and pipelines) to the new convention.

Despiking workflow without qc image and end-user script where the created workflow is used.

add minimalistic dockerfile which creates a container with functional nipype and all currnetly needed software
(will be extended continously)

consider using neurodocker: https://github.com/ReproNim/neurodocker

We need the following dirs:

examples: example pipelines (the current mapnode and the iterable pipelines must be moved here), mostly for "education" purposes, or to be looked up during development
pipelines: here we will have a collection of the actual PUMI-based pipelines for "production" purposes (like RPN-signature)
scripts: any helper scripts, + later maybe a wrapper for the docker command that runs pumi-pipelines

Necessary for the integration of the RPN-signature

nodenames and labelnames should have the same name.

Example: bet = pe.Node(interface=fsl.BET(), name='bet')

Do not do this until the whole old PUMI has been ported

Refactor the name of info_get.py to info.py.

There are some problems with the Dockerfile, especially if you want to run PUMI in PyCharm with a docker remote interpreter.

No longer use the latest software versions for FSL etc., instead use the versions with which we have tested PUMI.

Missing:

• Templateflow
• ANTs
• HD-Bet
• GPU support

Necessary for the integration of the RPN-signature

Run the bet_subworkflow on functional data and compare the results with the results of RPN-Signature

Notes on porting from the old PUMI version to the new one:
I will use the terms „NP“ for „new PUMI“ and „OP“ for „old PUMI“.

• Use NP NestedNodes and NestedWorkflow instead of nipype nodes and workflows
  • You only have to change the nipype-node-import to from PUMI.engine import NestedNode as Node and the nipype-workflow-import to from PUMI.engine import NestedWorkflow as Workflow
• Every workflow of NP must use a (suitable) decorator.
  • Anatomical related workflows use „AnatPipeline“, functional ones „FuncPipelines“ and QC-Workflows use „QcPipeline“
  • You have to specify the inputspec_fields and the outputspec_fields when using the decorator
  • It‘s possible to specify your own DataSinker regex-substitutions with regexp_sub=[]
  • You can deactivate the default regex substitutions with default_regexp_sub=False (not advised if you don‘t really know what you do)
  • Example usage with comments can seen here in https://github.com/pni-lab/PUMI/blob/main/PUMI/pipelines/anat/segmentation.py at the qc_bet definition
• Don‘t create workflow objects in your NP workflow methods. The decorator will create a workflow object and you can access it simply via using „wf“.
  • Example: wf.connect('inputspec', 'in_file', plot, 'registered_brain')
• Don‘t create inputspec, outputspec, and SinkDir objects in NP. The decorator will handle this.
  • You can utilize these in connect-statements by using the strings „inputspec“, „outputspec“ and „sinker“.
    • Example 1: wf.connect(bet, 'out_file', 'outputspec', 'out_file')
    • Example 2: wf.connect(bet, 'out_file', 'sinker', 'out_file')
  • The inputspec_fields that you specify in the decorator are the IdentityInterface fields of the inputspec object in OP
  • The outputspec_fields that you specify in the decorator are the IdentityInterface fields of the outputspec object in OP
• The first parameter of you workflow has to be „wf“ and the last one „**kwargs“. You do not need to pass wf yourself, the decorator will handle this
• Currently you have to pass "qc_dir=wf.qc_dir" when creating a qc object. We will fix this in the future. This can be seen in https://github.com/pni-lab/PUMI/blob/main/PUMI/pipelines/anat/segmentation.py in the bet_fsl workflow
• Don‘t use MapNodes. Most times it‘s enough to change MapNode to Node and delete the iterfields but verify that this makes sense when doing it!
• You may stumble across such connection statements in OP (note the „outputspec.[...]“ and „inputspec.[...]“): wf_mc.connect(mytmpfilt, 'outputspec.func_tmplfilt', fmri_qc_mc_dspk_nuis_bpf, 'inputspec.func').Don‘t explicitly mention inputspec/outputspec in NP when connecting nested workflows. The NP equivalent would be: wf_mc.connect(mytmpfilt, 'func_tmplfilt', fmri_qc_mc_dspk_nuis_bpf, 'func').

Also note the coding conventions mentioned in NP README (https://github.com/pni-lab/PUMI)

Necessary for the integration of the RPN-signature

A AnatPUMI pipeline that uses pydeface or similar to de-face an anatomical image.

• get in touch with Balint to discuss which de-face software to use

• create QC image (bet_wf)

• create a test