Hello Lukas:

As you stated, the only reason for using the "cliplevel" is to prevent loss
of numerical accuracy due to voxels with baseline close to zero. You should
set this "cliplevel" value low enough so that it does not exclude brain voxels.

The reason for loss of numerical accuracy is that, usually, the results are
stored as scaled short integers. The scale factor is determined by the
largest value (voxel) in the sub-brick. It is this scale factor which limits
the accuracy of the representation of the data. If a particular voxel's
baseline is close to zero, then division by the baseline may yield a very
large number, hence a large scale factor. Perhaps this (baseline near zero)
is not a problem for your data, in which case the "cliplevel" is not needed.

Yes, using a mask would be an acceptable alternative. But then, you would
have to determine the "cliplevel" for generating the mask.

Doug Ward