Hi,

A technical detail is that version 4 of this atlas contains 41 232 144 segmented voxels, which in the extreme case could become a set of point clouds (of the 222 segmented structures) totaling in the same amount of points, so ~41 million. How and if it should be downsampled in a random or uniform matter, largely depends on what amount of points the targeted VR environment(s) can handle. Assuming that I understand the question correctly and the plan is to have point clouds for the structures themselves (I'm not very familiar with VR and when I saw VR, that was working with meshes).

For simplifying the atlas itself, it's probably advisable to do that when the atlas is still NIfTI, then ITK snap or external tools can help with merging structures into larger ones. Merging them when they are subsampled point clouds already may easily result in unwanted artifacts, such as visible boundaries between supposedly continuous regions. A quick search revealed these ideas for such relabeling task: https://discourse.itk.org/t/remove-multiple-labels-in-1-bone-and-have-only-1-label-per-bone/5186

A side note about MeshView: it doesn't need Flash for quite some time (6-7 years). https://meshview.apps.hbp.eu/?atlas=WHS_SD_Rat_v4_39um shows WHS SD Rat v4 - however as it's not a VR tool, chances are that it can't help with your use case.

Best regards,
Gergely