This sounds to me like the same issue as in this recent thread. The problem is probably simply that you’re getting totally wild ADC values (i.e. values for the tensor elements) in non-brain regions. You try simply masking the data at the dwi2tensor stage, i.e.:
dwi2tensor dwi.mif -mask mask.mif dt.mif
That ought to get rid of most of the problem – but probably not all of it, there’s likely to be a few remaining non-brain regions included in the mask that still cause trouble…