add further atlases or keep it lightweight? [enhancement] about abagen HOT 5 CLOSED

@illdopejake is currently working to standardize nilearn fetchers, which might help here !

from abagen.

Hey @PeerHerholz! Thanks so much for reaching out with the suggestion. 🙂

This is something that I've been thinking about for various projects since it came up in miykael/atlasreader#4, and I unfortunately I have yet to settle on a good solution. I'd love to be able to use the atlas fetchers in nilearn, but unfortunately the metadata provided with the atlases aren't consistent enough to rely on. It would be great if something like HAWG could be adopted to standardize the metadata, but we're not there yet.

So my question in response to your question is: how would you decide which atlases to include? I have trouble settling on a relatively concise list, myself, and really only included the DK atlas because it's convenient (low resolution = quick to test!). I would be alright including a small list (maybe <5 atlases) I think, but worry that this is something that could quickly balloon out of proportion... Let me know your thoughts!

from abagen.

As @PeerHerholz sais, the question is lightweight or not. A nilearn fetch_ approach is probably possible, the metadata are not too chaotic and I'm sure that a PR to nilearn could take care of any issues.

But as I mention here (miykael/atlasreader#60 (comment)), I see some other issues with using nilearns fetch_ approach. The main one is the problem of file size. nilearn is prone to download much more than you need. But also this could perhaps been taken care of with a PR to nilearn.

from abagen.

Thanks for your thoughts on this @rmarkello and @miykael.

True that: deciding on atlases to include is rough. I had the same "problem" with ALPACA, but it actually also helped to get some ideas and plans. I think, that the total number should be restricted to around 5 atlases. Furthermore, these atlases should be a nice mixture of type (cytoarchitecture, function, multimodal, etc.) and complexity (small/few vs big/many ROIs).

Would datalad be an alternative to nilearn's fetch_ approach?

Another option could be to limit the number of already included atlases, while giving users the possibility to use the package with their atlas of choice, e.g., a small functionality and tutorial?

from abagen.

I'm going to close this since I think at the current juncture including atlases is outside the scope of abagen. Will maybe revisit at a later point but the plan is to stick to just processing and workflows, with the DK atlas included purely for example purposes.

from abagen.