Daniel,
 The issue you are experiencing reflects the fact that your image has the "Label" intent code, which suggests that the different values reflect discrete regions rather than scalar values. For example, consider the partial volume effect of a voxel that contains a proportion of Brodmann Area 19 and some portion of area 17 - in this case it would be incorrect to infer that this voxel is area 18. You can fix the issue by opening your header in MRIcroGL and setting the intention to "Not statistics". Your masking operation stripped the intention code, so that changed MRIcroGL's behavior to treat the new image as scalar values.

fslhd CombLesionMap
...
intent Label index
intent_code 1002