Byte swapping does not appear appropriate for this data,
as the negative values look correct (to me).

Without scaling, the data range is from -6464 to 32767.
The scale factor is 1.13753.

So applying AFNI_DICOM_RESCALE looks incorrect, or at least
unmanageable. The positive range of short integers already
hits the maximum signed short. So scaling by more than 1.0
will result in wrap-around, messing up your data.

I suggest that you do not use AFNI_DICOM_RESCALE. And if
the negatives in the data bother you, chop them off.

If you really want to scale the data, apply it after to3d
using 3dcalc, but writing the data out as float:

3dcalc -a anat+orig -expr 'a*1.13753' -prefix anat.scaled -datum float

- rick