Long-form output files

[mathesong]

Hiya,

As we discussed in my email, I'd love to know your thoughts about changing up the structure of some of the outputs. I think it boils down to your opinions on each of the following ideas, in order of progressively being closer to how the files will be used in later steps and further from making clear what was done to generate them in earlier steps.

1. Long-format with predictable column names vs wide format with variable column names.
2. Including all output information (e.g. volume, region index, TAC) in a single long output file, vs than multiple files.
3. Combining information across multiple segmentations into one long output file, vs one output for each segmentation.

From my perspective, all three make later analysis of the data easier. But 1 and 2, without 3, is almost equilvalent, but makes it easier to look over the separated outputs more easily. In fact, if people will run multiple pipelines - e.g. if one pipeline specially extracts midbrain ROIs, and another pipeline does the rest, then it might still be necessary to be combining tac outputs from multiple files regardless. So, for me, 3 is slightly less desirable than 1 and 2. But it all comes down to priorities and what the outputs should represent.

If you let me know how you prefer it, I could submit a PR.

/G

[mnoergaard]

Hi Granville,

Thanks a lot for these comments! I have been thinking a lot about this, and as I start to formalize my thinking with code and outputs, my thoughts/reflections are the following:

1. The definition of regions of interest need to be specified as a mask (dseg.nii) with a given voxel belonging to a certain index with a given label. The connection between the nifti file and the index id/labels need to go into a tsv file, where it is clearly specified how a given index label corresponds to a given brain region. This will make it clear when creating a *.tacs file and *.kinpar file what column name belongs to what region.

2. Ideally, I would like all regions to go into one file, but this is only possible when you have non-overlapping regions. For example, I focus a lot on the raphe these days, but the raphe is a part of the brainstem, so it would not be possible to put this into a single dseg file. Therefore, if overlapping regions exist, more files are needed to accommodate this.

3. Regarding volume of ROIs: because the dimensions are different than the tacs file, I suggest these go into a separate file (e.g. _volumes.tsv), specifying the volume of a given region belonging to a given dseg.

So I generally prefer it as specified here: https://www.dropbox.com/scl/fo/u7ose7r2fyuh4fqy93x7x/h?dl=0&rlkey=8prgqax8ve7h418l7il0wesl8

The extention _stats.tsv needs to be changed to e.g. _volumes.tsv.

Then everything else is in line with your suggested changes to the BIDS-derivatives spec.