I am curating a large collection of clinical data. Some of the images fall outside the current BIDS specification and I was hoping for some human advice to supplement my searches the the specification and the various BEPs.
Clinical diffusion-weighted images
Our studies often include a simple diffusion scan producing ADC, FA and TRACEW maps. Sometimes the original diffusion images aren’t even kept. These would be used as anatomical images, practically, for things like outlining ischaemic lesions. How should they be labelled. I was thinking:
I realize technically these are derivatives, but since they are produced at the scanner, we can’t describe a pipeline. In many cases there will be no raw image to point back to. So I think it’s best to treat them as raw data. It may be unnecessary, but it may help to use a filename entity to group ADC, FA etc from the same scan (it’s not always clear which are related, but sometimes the Series Number hint at this, e.g. 401, 402, 403 all derived from DWI in series 4, which is not kept).
I can’t see anything about CT in the BIDS specification. We’ll use these again for anatomy, so I’m thinking for my purposes I can get away with:
Hi, there is some work on representing CT in BIDS. There is an existing discussion over here, and I believe it has spilled onto the BEP (also linked): CT in BIDS format
Thanks. I was wondering how to find the BEP Google docs, because the Github page only has BEPS 001, 016 and 021.
Any thoughts about the clinical diffusion images?
I’m sorry, my eyes skipped right over the diffusion bit.
It’s definitely an arbitrary line, as even reconstructed images are “derivative”, but basically the policy has been that, if a data type is “typically” not the primary result of the scanner, then it will be described in derivatives, and require derivative metadata. It should be noted that derivative metadata can be quite sparse, if there isn’t much to say. For example, source dataset and file information is RECOMMENDED, not REQUIRED.
You basically have three options:
Create a raw dataset and, within a
derivatives/ subdirectory, include your ADC, FA and TRACEW maps. This is where fMRIPrep would look for lesion masks, as masks are derivative types.
Create your initial dataset as a derivative dataset. It is within the rules for raw data to be included inside a derivative dataset. The constraint is that, if it has a name that would be valid in a raw dataset, it must be a raw file, which will be satisfied in this case.
I would probably suggest option 2, now that derivatives are officially defined. The validator needs some work to catch up, but I think it’s more coherent in general to think of all datasets as derivative, and document their provenance, rather than cordoning off “raw” and “derivative” components of the data.