CIT168 node count

[andreifoldes]

Summary

I realize its rude to ask question about "dev" features, but when opening the atlas-4S152Parcels_dseg.tsv I noticed that there are 28 (lateralized) nodes, but in the original CIT168 there are 20 (bilateral) nodes: https://neurovault.org/collections/3145/.

Is this because they get squished in MNI template space?

[tsalo]

I believe so. I think @mattcieslak handled the initial resampling, so he could speak to that, but it makes sense that we would lose some regions when downsampling to 2 mm3 and warping to MNI152NLin6Asym.

[mattcieslak]

In the AtlasPack repo you can see all the steps that were taken to make these atlases. I don't think, particularly in BOLD, that averaging signal from analogous regions in different hemispheres is what most people intend to do. This script is where the splitting and renaming happens.

There are also lines in that script where you can see that some tiny structures are merged together to prevent the downsample-disappearance that can happen when you go to 2 or 3 mm. The names in the sidecars tell you what the original CIT168 regions were and how they're been combined.

[andreifoldes]

In the AtlasPack repo you can see all the steps that were taken to make these atlases. I don't think, particularly in BOLD, that averaging signal from analogous regions in different hemispheres is what most people intend to do. This script is where the splitting and renaming happens.

Right! I was trying to understand why there are 14 nodes instead of 20 shown in neurovault.org/collections/3145.

There are also lines in that script where you can see that some tiny structures are merged together to prevent the downsample-disappearance that can happen when you go to 2 or 3 mm. The names in the sidecars tell you what the original CIT168 regions were and how they're been combined.

Yes, thank you for linking that. I can see that node 10 and 11 are merged into node 7, which explains the difference. Would it be possible for XCP_D to consider the resolution of the provided functional files and if sufficient apply the "original" CIT168? If I understand correctly the original atlas was designed with 0.7mm anatomical and was later upsampled to 1mm in template space.
This would be very useful for using XCP_D with 7T files.

[tsalo]

Would it be possible for XCP_D to consider the resolution of the provided functional files and if sufficient apply the "original" CIT168?

I don't believe the other atlases that compose 4S are available at a higher resolution.

[andreifoldes]

I see! However for the atlases like the Tian or the currently discussed CIT168 this "resolution-aware" functionality would be useful no?

[tsalo]

In the case of individual atlases used by XCP-D, I agree that having the highest-resolution version as the "base atlas" is best. However, CIT168 is only used in XCP-D through the combined 4S atlas, so the overall atlas resolution is dependent on the lowest-resolution component atlas in that case.

[tsalo]

If XCP-D isn't using the highest-resolution (NIfTI) version of any of its other atlases, please let me know. I think I tried to grab the best version of each, but I might have missed something.