This seems to be an issue with how you are thresholding your images. The input images have the intensity range -0.1289..0.2294 (left) and -0.1453..0.2361 (right). The masks you created appear to be forced to be binary (0 or 1). The discontinuous values will be particularly influenced based on your smoothing setting for interpolation. This issue is described in depth here:
  https://www.nitrc.org/plugins/mwiki/inde...

I think you have a couple solutions. First, you want to preserve the values in your NIfTI images. It is hard to give a precise solution without seeing your images, but this solution looks useful:
  https://github.com/rordenlab/spmScripts/...
the other alternative is to save each mesh as mz3 and then use Matlab to mask the vertex colors. That may be a nicer solution, but it is more complicated.