Ongoing collaboration with AFNI developers on brain atlases

[niksirbi]

On 2024-01-18 we met with Paul Taylor and Daniel Glen of the Scientific and Statistical Computing Core at NIMH. Paul and Daniel are among the core developers and maintainers of AFNI - a leading software suite for analysing and visualising multiple MRI modalities.

During our meeting, we identified several potential avenues for collaboration, centred around brain templates and atlases.
Here are some ideas that came out of the meeting.

1. We will try to convert some of AFNI's nonhuman templates and atlases into BrainGlove Atlas API format and scope out how hard the task is and how we can make it easier. The macaque NMT v2 template + CHARM annotation could be a good starting point because I am quite familiar with it, and because we currently have no macaque atlas in BrainGlobe.
2. Daniel will explore doing the inverse, i.e. converting one of BrainGlobe's atlases to be used with AFNI.
3. Following the above, we can meet again and discuss what can be improved on our side, especially looking ahead to BrainGlove Atlas API v2.
4. @alessandrofelder is interested in atlas validation - i.e. a "checklist" to verify that an atlas is suitably constructed and packaged. Paul has thought about writing an article on that topic.
5. We are interested in building new brain templates, and we could get feedback from Daniel and Paul, to make sure we have done things properly, and to guarantee that these templates would be usable in AFNI and similar MRI software.
6. Related to point 5, Daniel mentioned that AFNI has a tool underway for constructing templates via iterative registration and averaging. He will share it as soon as some bugs are ironed out.

Feel free to add any points I might have overlooked @adamltyson and @alessandrofelder.

[afni-dglen]

Some tips on AFNI atlases -

Making them is described here:
https://afni.nimh.nih.gov/pub/dist/doc/htmldoc/template_atlas/howto.html

The atlas_point_list attribute for any particular atlas can be fetched on the command line with

3dinfo -atlas_points myatlas.nii.gz

Some items that aren't described in the previous link:

longname support (abbreviated acronyms are typically used for the regular region names)
representative coordinate (COG - center of gravity/mass is typically used, but we can choose all kinds of centers or any representative coordinate)

<ATLAS_POINT
 data_type="atlas_point"
 STRUCT="BSTLj"
 VAL="201"
 LONGNAME="lateral bed nucleus of stria terminalis,juxtacapsular subdivision"
 OKEY="201"
 GYoAR="0"
 COG="2.25 0.15 0.6"
/>

Probabilistic atlases are supported via an SB_LABEL field (similar to the STRUCT field) for the subbrick volume within the series of volumes in a dataset file.

The atlases can also be configured to specify:
web links for regions
web links for region connections