Hey Dirk,
  See the response below:

On Mon, Dec 8, 2008 at 9:44 AM, Dirk Reiners <dirk@lite3d.com> wrote:

Basic question: how are you doing the volume rendering? Texture slicing or shader-based ray marching?

I'm simply doing texture slicing for volume rendering.

You can avoid the voxelisation by not storing a bit but a plane equation in the voxel. That would allow fairly high precision clipping, even for an 4x8 bit texture. But it really is only necessary if your volumes are small, so that a voxel is blown to many pixels. Is that the case?

I am running with low resolution textures (10x8x8 to 20x16x16 or so).  I'm not sure I understand the full implementation of storing a plane equation in the voxel, can you elaborate a bit more?

  3. Use a shader to clip the ROI's

That sounds like the best idea, IMHO, as long as you have a relatively small number of ROIs. How many are you expecting?

I'm expecting on the order of 4-6 ROI's per dataset and may have 2-4 datasets up at the same time.

I've pinged the opengl groups too, they recommended using using the multitexture extension with an alpha texture with higher resolution than the dataset to minimize the voxelation.  I think this is pretty interesting too.