Hi Georg 

That would be the best way to get to your results.

If you use StatNonParamTestPDM on the difference magnitude files, then you should be fine too i.e. the results are meaningful, you just cannot covary for anything, such as gender, age, etc, though I am not sure you want to do that in your case (as you are mainly looking at measuring the local error between your manual and automatic segmentations) .

You can use R, like we did. FSL's or Freesurfer's tools should work too. To reformat the results in a format to be use with MeshMath -KWMtoPolyData is simply. The result file should look as follows:
NUMBER_OF_POINTS=
DIMENSION=1
TYPE=Scalar


Hope this helps
Martin