Hi gajan,

Actually I work in medical imaging myself, and this project was a result of applying what I learnt there to the field of gaming. So I think there is a good chance you can make use of the library.

I think this should be fine. I have loaded a volume of this size and actually the frame rate was a little low (maybe 10 fps) but that was on an oldish computer (Athlon 1800+, GeForce 6600). Also, you main concern will probably not be the size of the volume but the number of triangles generated. These two things are related of course, but are not the same. It was the graphics card which was the limit, not the amount of memory. Also, I have made a lot of changes since then as it was 18 months ago. I think you will be able to render the scene fine, but see my note later about the physics.

This could be a sticking point - in my library each voxel is a value between 0-255 which represents the material. If you wanted to load a CT scan you would have to define a few materials such as bone, tissue, fat, etc, and then convert to these. For example, you could convert any voxel over 1000HU to the 'bone' material. So you will have to do some classification, but then you can have pretty, texture mapped bones which cast shadows, etc

Well I think that's really just a matter of how you interpret/visualise it. Not sure what to say here, except that I use the marching cubes algorithm so make sure you understand what that outputs.

My project is broken into two parts - the PolyVox library and the Thermite game engine. The PolyVox library is only concerned with generating the mesh, it does not handle any physics or rendering. The physics is implemented in the Thermite game engine which in turn uses the 'Bullet' physics engine. I think you would have three options here:

1. Use the whole game engine. Thermite would do the physics and collision detection for you (via bullet) but you wouldn't really have any control. I guess this is not the best option.
2. Use just the PolyVox library and implement your own physics based on the mesh which is given to you. You could write your own algorithms, or integrate an existing collision detection library.
3. You have access to the raw voxel data, so you could implement your own collision detection based on this. Collision detection against voxels is very easy because you just have to check whether a voxel is occupied or not.

Note: In the case of 1 (and possibly 2) you might find you struggle with the sheer number of polygons. In Thermite I am actually using a lower resolution mesh for the collision detection. PolyVox can provide this for you.

You can modify the voxel values, but this changes the material (not the density). See point 2. You can remove material by setting the value to zero.

Basically yes, but see also point 4.

Make sure you try the demo from the homepage - it's only a 256x256x256 volume but should give you an idea of the engines capabilities.

Hope that helps!