Derivatives for source reconstruction data (EEG/MEG)

[christian-oreilly]

I am currently working on a grant proposal that would involve, among other things, a better support EEG/MEG sources as derivatives. My focus would be to implement the support for dealing with such derivatives in MNE-BIDS (mne-tools/mne-bids#1176), but I think it may require some work on BIDS itself to ensure that the way we store these data is standardized and codified in BIDS. I am not quite sure what is the best way to move forward on this (just a standalone issue, an independent BEP, as part of an existing BEP/issue). I'd welcome some guidance on how you think this topic can best be organized within the existing structure.

[Remi-Gau]

We usually try to ask people to first look at rhe exiating list of BEP to see if there is not one that already match what they want to do.

https://bids.neuroimaging.io/get_involved.html#extending-the-bids-specification

I think that in your case you want to have a look at the extension 21.

I think that @dorahermes @sappelhoff @guiomar @CPernet @robertoostenveld may be able to tell you more on the best way to help them on this front.

[CPernet]

For this type of derivative, there are two parallel extensions to follow, the BEP21 as Remi mentioned and the general derivatives (more like a series of PR) for common data format.

It obviously depends what is the project - note that current spec allows storing gifti files and thus source reconstructed data - there is a current PR for the agreed general format for hd5 and zarr for N-dimensional arrays if you need more than 4D. There are also several BEP for connectivity if that's what you plan to do.

Last but not least, BIDS aims at being inclusive, and therefore supporting multiple ways to do something ... ie any source reconstruction stuff got to work no matter if one uses MNE, Brainstorm, FieldTrip or EEGLAB.

[sappelhoff]

cc @christinerogers as well, as I saw EEGNET mentioned.

[christian-oreilly]

I will take some more time to look into BEP21 and the other information provided. Thanks for the info. I am not sure if general 4D storage will work. In a different project, I had a collaborator who was more comfortable with fMRI data so he wanted me to transform EEG sources into nii files that he could visualize with his usual tools. This did not work so well because of the differences in spatial and temporal resolution in EEG vs fMRI. It really depends on how flexible the BIDS definition for 4D storage is (e.g., voxels versus vertices). I'll check it out.

Indeed. On the software side, I'll implement it in MNE (and I plan to work with EEGNet too as mentioned by @sappelhoff), but any adjustment for BIDS would be software-independent.

[christian-oreilly]

I think that BEP21 was adopted a while back (~2021) and is currently not being worked on, is that right @sappelhoff? What is the policy about approved BEPs? Are they still open for submission and approval of new versions or are such modifications made through new BEPs? I'll go through BEP21 doc and see how it supports the use cases I have in mind, but before I dig into that, it would be useful to have more clarity on the best way for me to propose changes/updates if any were required.

[sappelhoff]

BEP021 was dormant for a while but is actually experiencing a revival since a few months, check also: https://github.com/bids-standard/bep021/issues and have a look at recent issues.

and yes -- comments are welcome everywhere.