BIDS wishlist for nipype 2.0

[adelavega]

Starting this issue as a place to gather discussion about BIDS-awareness we'd like to include in nipype 2.0 (most of these suggestions came from the nipype 2.0 sprint).

Domain specify versions of these interfaces already exist in existing projects such as fitlins, fmriprep, mriqc and more. It would be great to abstract these out into general purpose Interfaces.

• Improvements to BIDSDataGrabber

  • Update for pybids 0.7, correct defunct BIDS-entities (e.g. 'modality')
  • Enhanced iterables. Currently, iterables are given as a list of already known items (e.g. subjects you want to iterate over). It would be nice to be able to say you want to iterate over subjects, without knowing the subject IDs before hand.
  • Could also read from a BIDS-model and iterate over relevant subjects/runs/tasks

• BIDS metadata awareness in interfaces

  • Execution interfaces are BIDS-aware and pull out relevant meta-data from BIDS project, such as TR
  • Interfaces pass along entities associated with outputs. If output paths are consistent but not BIDS compliant, could write converters (e.g. converting FSL outputs to BIDS)

• Derivatives compliant BIDSDataSink

  • Writing DataSink substitutions is one of the biggest issues for beginners. It would be great if you could plug the outputs of Interfaces into a BIDSDataSink and output a valid BIDS derivative dataset.
  • BIDSLayout can be used to write BIDS paths, but the challenge is obtaining the entities for outputs. At first, it may be good enough and useful to let users manually specify these (if they know them), and just let the BIDSDataSink handle formatting / dataset creation.
  • If previous interfaces are BIDS-aware, entities will be known and can be used to generate paths.

• BIDS Specify Model

  • Pybids Analysis wrapper, which can be used to specify models for FSL, SPM, nistats from a BIDS-Model.
  • Related issues: Simplified/alternative way to specify models #1359 New SpecifyEvents interface #1555

Of course, none of this would be complete without docs, so once this is release it would be great to include it in the niype tutorial.

Other suggestions are welcome.

[hbraunDSP]

Seconding the need for a BIDSDataSink. After a couple hours looking for a better solution I'm resorting to putting a Rename node before the DataSink for every single file. For those of us trying to comply with BIDS this is something that should be routine and easy, but isn't.

I see there's a BIDSDataSink interface class in https://github.com/poldracklab/fitlins . Is this ever going to be added to nipype?

[effigies]

@hbraunDSP There are also derivatives data sinks in https://poldracklab.github.io/niworkflows/api/niworkflows.interfaces.bids.html#niworkflows.interfaces.bids.DerivativesDataSink. We would need to agree on the desired features to bring something into nipype.

It may also make sense to have small interface packages that allow projects to select which interfaces they want, which would make it a bit less of a fraught decision, in terms of maintaining backwards compatibility.

Anyway, the niworkflows interface has been used in several projects, while the FitLins one is only in one, so you might see if that does what you need.