search

SCIInstitute / ShapeWorks

ShapeWorks
http://sciinstitute.github.io/ShapeWorks/
Other
104 stars 32 forks source link

Volumetric correspondence models #989

Open sheryjoe opened 3 years ago

sheryjoe commented 3 years ago

Related to Andy's project.

Use multidomain, one of the bone surface (for surface correspondence) and one for the imaging data (for interior correspondences). Use the bone surface as a free-form constraint for the image data and disable the surface constraint for the image data.

sheryjoe commented 3 years ago

Related to #1182

akenmorris commented 3 years ago

So we will use image data as an attribute? Will we assume that the volume data has already been intelligently normalized somehow (e.g. MRI?).

@HeavenlyBerserker , is it easy to specify a mesh as a free form constraint?

HeavenlyBerserker commented 3 years ago

I don't understand this issue. What are interior correspondances? I'm also not sure what you mean by "Use the bone surface as a free-form constraint for the image data".

FFCs are stored as meshes. I turn image data into meshes and define FFCs using mesh vertices and faces.

akenmorris commented 3 years ago

In this case, we will use volume image data such as a CT as the domain. Particles will not be restricted sit on the surface, but instead must remain inside the surface. Otherwise, they are free to move around inside the surface, but will maintain correspondence with other shapes using x/y/z and volume data.

We need some way of constraining that the particles sit inside the surface. Does that make sense?

I think what you are working on will work for this. What's the current state?

HeavenlyBerserker commented 3 years ago

Ah, I see. Yeah, should work with some tweaks.

It's going well, but it still does not work 100%. Still debugging.

HeavenlyBerserker commented 3 years ago

Wait, how will you get volumetric queries from a mesh?

akenmorris commented 3 years ago

The domain is a 3d image volume, not a mesh. We only have the surface as a constraint.

sheryjoe commented 3 years ago

@HeavenlyBerserker this is where RBF could help to get a constraint field (h(x)) defined on the 3d volumetric image domain/space. The field should encode signed distance to the surface. A cheaper alternative would be to use a signed distance map for the surface to define in/out of the surface and the map gradient can direct particles move away or toward the surface. One potential caveat is if we have a complex surface with thin structures and high curvature regions (e.g. pelvis).

HeavenlyBerserker commented 3 years ago

I see, yep. Meshes wouldn't be enough for this. RBFs should work, I suppose.