Is there a way to detect feature points of volume(vtkImageData) like in mesh(vtkPolyData)?

Hi guys,
I’m trying to detect feature points from given input imagedata. More specifically, given CT scan of an object with visually-obvious concave and concave parts, return concave points and convex points.

The only way I can think of doing this is by:
1.convert imagedata to polydata using marching cube(given isovalue)
2.detect feature points using polydata. specifically, for concave/convex points, I can use curvature filter and easily find these feature points.

This is doable.The problem is that I lose precision during conversion from imagedata to polydata.
I don’t think I can calculate the curvature of imagedata, is it possible to detect concave/convex feature on imagedata itself?