External Email - Use Caution

Dear Dr. Greve & FreeSurfer developers,

I am comparing two FS-based ROI extraction strategies for a dynamic PET study ([18F]SynVesT-1, FreeSurfer 8.1.0), both using the Desikan–Killiany parcellation and an identical, volume-sampled centrum semiovale reference region. Regional binding potentials (BPnd, SRTM2) were derived from tacs without partial-volume correction using:

1. aparc+aseg volumetric labels (1 mm) in the T1 native space;

2. gtmseg labels (leveraging the gtmseg tool in PETsurfer), same DK parcels (no editions to the gtmseg.mgz).

Across the 68 cortical ROIs the two agree extremely well as expected, but gtmseg regional BPnd is systematically ~21% higher than aparc+aseg (and reported BPnd values in the lit using the same reference region) consistent across subjects and regions.

I understand these are the same parcellation scheme but not the same segmentation: gtmseg defines the cortical ribbon from the white/pial surfaces at 0.5 mm, whereas aparc+aseg uses the 1 mm voxel ribbon. My interpretation is that, without PVC, the surface-defined ribbon recovers purer? GM (less WM/CSF dilution), which would explain a positive offset in this direction.

But, my doubts:

1. Is a systematic ~20% higher regional mean from gtmseg vs aparc+aseg, under no-PVC extraction, of the expected magnitude in your experience, or does it point to something we may be doing wrong?

2. For no-PVC regional extraction specifically, which segmentation do you consider the more appropriate substrate, given that gtmseg is optimised as the input geometry for mri_gtmpvc rather than for direct sampling?

Thank you very much for any guidance and the great efforts put on surface-based methodologies!

Best wishes, Carme Uribe CAMH, Toronto