We are recently trying to get cortical thickness from freesurfer and converting them to 32k. A problem we notice is that after converting to 32k, all the cortical thickness values become negative. However, when we load freesurfer thickness files in freeview, everything appears normal. We wonder if merely inverting what we currently have in 32k would be a good idea. Alternatively, we also calculated the Euclidean distance of pial and white after they have been converted into 32k space, which should be cortical thickness. However, when comparing the inverted values and the “pial-white” values, we find that there are some differences. In this case, has anyone ever had similar issues and have suggestions on which one to trust?
FreeSurfer’s calculation of thickness is not the expected, simpler distance between white matter and pial surfaces; it’s actually a little more complicated. First, the computation is done between the smoothwm surface, not the default white matter surface, and the pial surface. For every corresponding node, the shortest distance of that node to any other node on the other surface is found. Those shortest distances from pial to smoothwm and from smoothwm to pial are then averaged to compute the thickness. Thicknesses are limited to a range between 0 and 5 mm. I’m not sure how you find negative values, but that shouldn’t happen with typical processing. There are, of course, other interesting ways to find thickness.
Hi, thanks so much! On that note, do you have any experience on how to convert freesurfer output to native space/32k/164k? If so, I would greatly appreciate it if you could point me to any sources. Thanks agian!
I’m not too familiar with the officially sanctioned FreeSurfer way. In AFNI, there is a procedure to convert the FreeSurfer surfaces (@SUMA_Make_Spec_FS) into standard meshes that maintain correspondence across subjects using FreeSurfer’s registered sphere warps.