Please excuse my delayed response, I've just moved to a different continent. :)
That's a fundamental image processing problem, and will be different for every type of data and even for every dataset. What has worked for me in the past, is to use a combination of RegionGrow and ImageMask to mask AWAY nearby bones, and then to use these same modules followed by Contour and WindowedSincSmooth (or something like that) to create good surfaces. In your case I would dive into the literature to see specifically what research has been done on the segmentation of vertebrae from CT data.
Good luck,
Charl