For nuisance regression the software will automatically select GLM rather than Liebermeister measure. Specifically, I would add a negative value to the number of permutation which will enable the Freedman-Lane method. In general, unless you have a very large sample, you may want to consider a region of interest approach, which provides more statistical power than the voxelwise approach. With regions of interest, the lesion score becomes continuous (e.g. proportion of region damaged).