Clarification on the documentation and open to collaboration

[jcohenadad]

Hi @benoitberanger,

Thank you for creating this BIDS conversion toolbox. It looks like a great tool! To answer the following point in the README:

There are several packages that have a semi-automatic approach to organize datasets to BIDS, such as Dcm2Bids However the objective here is to be 100% automatic.

I would like to kindly point out that dcm2bids is in fact fully automatic, as long as the configuration file covers the proper use cases, so to make it 100% compatible only requires to expand the existing configuration file to larger scenarios. Based on what I understand from how niix2bids works, it also uses an ad hoc configuration files (here and here) that needs to be tweaked for new sequences, scanner model and vendors.

Given that both niix2bids and dcm2bids are both open source and relatively small projects, I think our communities would benefit greatly from collaborating with each other, so there is less effort spent in maintaining several packages. I am not saying that niix2bids or dcm2bids projects should be discontinued, but I just would like to start a dialogue, to see how our teams can collaborate and how we can essentially merge our efforts.

[benoitberanger]

Hi @jcohenadad,

Thank you for your taking a look at this project !

Context

First I would like to explain a bit the context (I might add some info in the README later). I work in a lab acquiring data for many different protocols. We need to apply BIDS architecture to our homemade archive architecture, where every protocol have a different way to name their sequences. A "simple" 3DT1 can be called : T1w, 3DT1, anat, anatomy, anatomy_1mm, mprage, mprage_1mm, MPRAGE,.....

Use case

example 1

If I need to prepare BIDS data for a single (or a few) cohorts, I would use Dcm2Bids, since it does not take too much tome to prepare each config.json (or a signale one flexible enough). Also, Dcm2Bids does not take shortcuts like niix2bids does.

example 2

However, if I need to prepare BIDS data on heterogeneous cohorts, I would prefer to use niix2bids since I don't need to prepare a config.json. However, niix2bids takes shortcuts, even though it's bids-validator valid. For instance the IntendedFor field is completely ignored. On the other hand, when I need to perform quality control such as MRIQC on all BOLD data from all protocols in the last 5 years, I will obtain a BIDS architecture in just one single command, ready to perform the processing workflow.

So ?

My opinion is Dcm2Bids and niix2bids do not share exactly the same objective.

Limitations

Based on what I understand from how niix2bids works, it also uses an ad hoc configuration files (here and here) that needs to be tweaked for new sequences, scanner model and vendors.

You are right, for the moment niix2bids is limited to Siemens magnets. It's a work in progress, I only have enough data from my lab, which only have Siemens.

So far, scanner model do not matter. At least for Siemens, it's ok.

Sequence ? You have a good point. So far, most Siemens Product sequences and most common C2P (like CMRR MultiBand) are compatible and tested. I tested most of the data we have in our lab. To add new sequences, or a user can make a PR so we add it in the default config file, or he can modify his local version. But this is once and for all.

Collaboration & effort merge

100% agreed !

[jcohenadad]

Thank you for the thorough answer @benoitberanger. I'll give it some thoughts on how we could merge efforts. Hopefully the community can also chip in with ideas / initiatives.