How to use kaolin in CT?

[Monsoon-k]

Such amazing work! You mentioned that kaolin physics works well in CT Scanns such as Heterogenous Skull/Brain, would you kindly give a tutorial of that? I appreciate your help!

[shumash]

We would love to support this use case. For a visual example showing that it's possible, see Simplicits paper website, which uses the same physics algorithm and implementation as Kaolin.

@itsvismay please provide some pointers to get started.

[itsvismay]

Hi @Monsoon-k, for the CT simulations in our paper, I segmented (or thresholded) organs using their hounsfield density values.

For data I used scans from the visible human dataset. For organ segmentation I used either Paraview or Slicer 3D to strip the skull, brain or bladder. Here's an example of stripping a skull in Slicer (https://www.youtube.com/watch?v=GKo7TyboLsg).

I will add a video tutorial for simulating organs to my Todo list.

[itsvismay]

For more context here are the steps to simulate a skull/brain:

1. Find CT scan data from an online dataset. I used the visible human data.
2. Strip the skull from the brain in Slicer3D. You can find a tutorial online and save out a point cloud or even a high-res surface mesh for the separated skull and the brain. (One thing to note is that if pieces of stiff material are highly interspersed with soft material, then Simplicits will have trouble separating the deformations and it will cause locking. So make sure the soft brain region is unified and distinct from the stiff skull)
3. You can load these objects into simplicits as shown in the simplicits_low_level_api.ipynb. Sample various points and assign the appropriate YM to each sample point. Training and simulation should work the same as it does now.

[github-actions[bot]]

Stale issue message