Very hard to fix, because it was never in the original concept. SliceBasedSegmentation was/IS primarily a tool for the context of liver surgery planning, where CTs are used.
Since it has become a general-purpose segmentation tool by now, MRIs should also be supported, but it is unclear how rotated images should be handled. Many tools require (at the moment) to write back one slice of the segmentation (binary image). With rotated images it is impossible to determine the "current slice" from a hand drawn (or other) contour.

Even if this would be changed and some voxels "inside the contour" (hard definition!!) would be filled if the contour is rotated against the original image orientation, a problem with holes in the segmentation could arise. If voxels are filled within one rotated contour, then the user advances to "the next slice" (which is not an original image slice) and draws an identical contour, it should be guaranteed that the 3D binary image of the segmentation does not contain holes. This is not trivial to guarantee -- as far as I understand it. I'll attach a sketch illustrating the problem.

Illustration

revision 15917

Today I checked in some changes related to this bug. RenderingManager is now capable of 'snapping' the image to the next axis aligned view that roughly preserves the image orientation but takes out any rotation other than multiples of pi/2.

the issus "geometries of generated volume-meshes do not match the image any more." is probably not solved by these changes.

Ok, fixed as far as possible with the current design. Drawing on rotated views will not work in the next future.