Long Term Maintanence of BIDS dataset

Hi BIDS experts,
I am currently working with different BIDS datasets, and have several questions regarding the long term maintainence of BIDS dataset. Usually, I use heuDiCov to do the conversion/organization from dicom to nifti BIDS and do the preprocessing with fMRIPrep, etc. I am wondering how could we update the BIDS dataset when we have new images coming to the existing BIDS dataset, do we have tools for this purpose? For examle:

  1. What are the recommendations (practical guides) for adding/deleting or changing subjects (images of new subjects) for the existing BIDS dataset?
  2. What are the recommendations for adding/deleting or changing sessions of images of subjects (new sessions for the existing subjects) for the existing BIDS dataset?
    Great thanks!
    Vincent (Qing Wang)

Vincent (Qing Wang),

As I understand it, Datalad is a nice solution for keeping curated datasets up-to-date: The Handbook — The DataLad Handbook

I helped to curate the ABCD-BIDS Community Collection 3165 on NDA and that is the direction they are headed now. That said, there are some simple changes you can make in your data as-is to answer the questions you posed (I think):

  1. To add: Just add the subject/sessions and keep tabs of everything available within a participants.tsv file (with a paired sidecar participants.json data dictionary of course). To change or delete/remove: Change away, but add something in your modality-agnostic CHANGES file to indicate why you made that change or perhaps notate a change in your participants.tsv file explaining why.
  2. As new sessions come in, just like in #1 above here, add them in and update your CHANGES file and consider adding something to your README about available sessions across some/most/all subjects.

Altogether, I would say when curating BIDS data to just think about the end-user receiving or downloading your data. They are your “customer” and the person you want to know what you’ve curated.


Great thanks for your detailed answers and suggestions, will try to explore DataLad!