I am writing to the mailing list to get some confirmation about normalizing patients with chronic stroke lesions. I am following the procedure as described in the post:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;fe3abb88.1601 (i.e., setting the part of the image corresponding to the lesion to zero before putting the image in the "Segment" module). 

Here is a Dropbox link of an image I took looking at all of the tissue probability maps.
https://www.dropbox.com/s/qu6jnwqp4v4ibnc/segmentation_lesion_set_to_zero.png?dl=0
From right to left and top to bottom, c5,c4,c3,c2,c1. Without knowing the inner workings of the algorithm, I would have thought that it would simply ignore any voxel whose value is exactly zero. The lesion's probability has been set to be mostly to c5 and a bit on c4. The probability of the lesion in c3,c2,c1 is zero. Is this the desired effect of "lesion masking" as described in the previous post?

To compare, I used the "Old Segment" module which allows one to explicitly put in the lesion mask (ones everywhere, zeros inside the lesion) with the original image. The normalized image that comes out of "Old Segment" has a slightly larger lesion, than using the "New Segment" method and setting the lesion to zero. Could this mean that Old Segment has respected the lesion size better than New Segment module? The difference isn't that large, but a ring around the lesion is visible when subtracting the two images (see picture https://www.dropbox.com/s/oxencpuou156qdb/difference_new_minus_old_segment.png?dl=0).

If this procedure seems right to you, or if anyone has a different way of minimizing the lesion's effect on normalization, I'd appreciate any input.