External Email - Use Caution
Hello,
I am using PETSurfer/mri_gtmpvc for PET quantification in the WRAP Alzheimer's cohort. I am encountering a reproducible error when regions present in gtmseg.mgz have no voxels after the segmentation is mapped into PET space.
I searched the mailing-list archives and found a previous discussion of the same error message involving segid 29. In our data, however, the affected labels differ between subjects, so I wanted to ask what the recommended general approach is.
What I am trying to accomplish ------------------------------
We are processing motion-corrected, frame-averaged 11C-PiB PET images with PETSurfer.
Our PET images are approximately:
Matrix: 128 x 128 x 63 Voxel size: 2.574 x 2.574 x 2.425 mm Voxel volume: 16.07 mm^3
The corresponding gtmseg.mgz images have:
Voxel size: 0.5 x 0.5 x 0.5 mm Voxel volume: 0.125 mm^3
Command -------
The relevant mri_gtmpvc options generated by our PETSurfer workflow are:
mri_gtmpvc \ --psf 0 \ --no-tfe \ --seg gtmseg.mgz \ --default-seg-merge \ --auto-mask 10 0.01 \ --mgx 0.01 \ --no-reduce-fov \ --no-rescale \ --save-input \ ...
I have attached the full PETSurfer log from a representative failed subject. It contains the complete generated command and output.
Exact error -----------
For example:
ERROR: creating region for nthseg=22, segid=78, Left-WM-hypointensities It may be that there are no voxels for this seg when mapped into the input space.
I initially encountered this problem in another subject for segid 79 (Right-WM-hypointensities).
Before changing the segmentation or adding replacement rules, I tested four additional independent PiB subjects using the same unmodified workflow.
gtmseg completed successfully for all four subjects. PETSurfer completed successfully for one subject and failed during mri_gtmpvc for three subjects.
Results -------
The regions producing mri_gtmpvc errors were:
Subject SegID Region GTM vol PET vox ---------------------------------------------------------------- WRAP1 78 Left-WM-hypointensities 6 mm^3 0 WRAP1 79 Right-WM-hypointensities 5 mm^3 0
WRAP2 31 Left-choroid-plexus 6 mm^3 0 WRAP2 78 Left-WM-hypointensities 21 mm^3 0 WRAP2 79 Right-WM-hypointensities 1 mm^3 0
WRAP3 78 Left-WM-hypointensities 2 mm^3 0
We also have an informative within-subject comparison in WRAP3:
segid 78: GTM volume = 2 mm^3 PET-space representation = 0 voxels mri_gtmpvc reports an error for segid 78
segid 79: GTM volume = 46 mm^3 PET-space representation = 1 voxel mri_gtmpvc does not report an error for segid 79
For the failing labels, we directly counted the voxels in both gtmseg.mgz and the PET-space segmentation at:
no_pvc_6.0fwhm/aux/seg.nii.gz
In every case listed above, the label is present in gtmseg.mgz but has zero voxels in the PET-space segmentation.
This makes us suspect that the immediate failure occurs when a region retained in the GTM definition receives no voxel representation after mapping the segmentation onto the much coarser PET grid.
One detail that of note was that total region volume alone does not determine whether a label survives. For example, segid 78 in WRAP2 occupies 21 mm^3 in gtmseg.mgz, which is greater than the volume of one PET voxel (16.07 mm^3), but it still maps to zero PET voxels.
Questions ---------
1. What is the recommended PETSurfer/mri_gtmpvc approach when a valid region in gtmseg.mgz has zero voxels after mapping into PET space?
2. For segids 78 and 79 (Left/Right-WM-hypointensities), is --merge-hypos intended for this situation?
3. We also see the same problem for segid 31 (Left-choroid-plexus). Should choroid plexus be merged/replaced for GTM analysis, or should this be handled differently?
4. More generally, should very small regions be merged before GTM construction, or is there another recommended way to handle regions that disappear at PET resolution?
Additional FreeSurfer context -----------------------------
The anatomical FreeSurfer reconstructions were originally generated with FreeSurfer 8. The PETSurfer/gtmseg environment currently uses FreeSurfer 7.4.0.
Because the FreeSurfer 8 reconstructions did not contain the classical talairach.m3z expected by the FreeSurfer 7.4 gtmseg, we generated talairach.m3z using FreeSurfer 7.4 mri_ca_register.
gtmseg subsequently completes successfully for all subjects tested. The error described above occurs later during mri_gtmpvc.
1) FreeSurfer version: freesurfer-linux-ubuntu20_x86_64-7.4.0-20230510-e558e6e
2) Platform: Ubuntu Linux, x86_64
3) uname -a: Linux hpclogin.sri.utoronto.ca 5.4.0-132-generic #148-Ubuntu SMP Mon Oct 17 16:02:06 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux
4) PETSurfer log: See attached.
I have attached the PETSurfer log from WRAP2, which demonstrates the failure for three different labels (31, 78, and 79).
Thank you,
Qassim Alkassir