If this is it and not working, I wonder that editing the aseg to exclude the nonlesional GM outside the pia, but leaving/ adding lesional GM of interest, either under the GM label or a custom label would allow create a 'proper' volume, followed by surface generation. I known this is doable, the question is whether the generated surface can be registered to sphere and whether DK atlas can be applied (this is a more general question, whether any surface generated outside the default recon-all pathway can be registered to sphere and in effect become "pia" or another classical boundary surface?)

example after labeling the lesion as 'o' (other):

% generate mask from aseg
cd /usr/local/fs5.3/freesurfer/subjects/subjid
mris_fill -c -r 0.1 surf/rh.pial mri/rh.pial.filled.mgz && \
mris_fill -c -r 0.1 surf/lh.pial mri/lh.pial.filled.mgz && \
mri_concat --combine --i mri/rh.pial.filled.mgz --i mri/lh.pial.filled.mgz --o mri/pial.filled.mgz && \
cd ./mri && \
mri_binarize --i aseg.mgz --match o  --o lesion.mask.mgz && \
fscalc pial.filled.mgz add lesion.mask.mgz -o pialmask.mgz && \
mri_convert brain.mgz brain.float.mgz -odt float && \ 
mri_mask brain.float.mgz pialmask.mgz pialbrain.mgz

% generate new 'pial' surface:

mri_tessellate mri/pialbrain.mgz -a surf/lh.pialbrain.pial

Please advise,

Octavian