On inspection of bias output files, we sometimes see spurious values
cropping up. For example, in a typical bias file, we see a range of numbers
from say 1.75 to 2.25 or so. Then there will be just a few values at 2.5 or
3.2 or so.
Per slice, there may be none of these odd values, or maybe 2 or 3 such
pixels, generally located at the border between brain and non-brain (ie:
bordering on the non-brain region that has zeros in the input volume, but
still actually in the non-zero region.).
Just on the basis of smoothness, it seems to me these values should not be
occurring... any clues as to what's going on?
When inspecting the actual output seg volume, we see that the input values
have indeed been multiplied by these out-of-whack multipliers.
We have several MRIs that exhibit this phenomenon.
Using FSL 3.3 on linux.