Export the vertebral labels as 3D models positioned near the corresponding spine segmentation for visualization

[miritrope]

Hi,
I’m using TotalSpineSeg and I see that the preview images show vertebral labels (L1–L5, T, C, etc.) positioned near the segmented spine levels.

Is there a way to export these labels as actual 3D models or objects that can be loaded alongside the spine segmentation? Ideally, these label models would be spatially positioned near their respective vertebrae in 3D space, for use in visualization software.

Thanks in advance for any guidance!

[NathanMolinier]

Hi @miritrope, thank you for testing TotalSpineSeg! Unfortunately, this feature hasn’t been implemented, as we typically rely on voxel values along with the class mapping provided here for downstream processing. I’m also not very familiar with approaches that embed labels directly within the segmentation space, how would reorientation or resampling be handled in that case? If this feature is important for your project, I’d be happy to explore possible solutions together.

Related issue

#106

[miritrope]

Hi @NathanMolinier, thank you for the quick and thoughtful response!

The reason I'm asking is that I'm working on visualizing spine segmentation results in 3D, and having the vertebral level labels (e.g., L1–L5, T, C) as 3D text objects or meshes positioned near their corresponding vertebrae would make it much easier to interpret and present the output — especially in more intuitive, spatial formats.

I completely understand that the system currently uses voxel values and class mappings for downstream processing. What I’m looking for is a way to extract or generate label metadata (e.g., vertebral centroids or landmark coordinates) that could be used to position 3D text in space — similar to what’s shown in the preview images, but as reusable 3D elements.

If there’s a way to hook into the current pipeline to extract this spatial label information, I’d be happy to explore or contribute ideas toward that. It would be a helpful feature for making the output more accessible and interpretable.

Thanks again,
Miri

[NathanMolinier]

I totally see how this feature could be useful, but it's not part of my immediate plans. If it's something you really need, I’d encourage you to contribute by opening a pull request. Once you have a prototype, I’d be happy to help with the full integration into the codebase. Thanks for your understanding!