Hi Bruce,  This would be most helpful. And yes, I think multiple labels for each voxel that are the coefficients that explain the observed intensity of that voxel would be a good format. I am thinking that anyone could then correct for pve of structures taken from the aseg.mgz file in matlab (or python) by keeping or removing voxels on the border of the structure based on which tissue class as the highest coefficient. Does this seem reasonable to you? Is that how the --pv flag for mri_segstats works (i.e., by making the decision to keep or discard a voxel based on the tissue class with the highest coefficient)?