Yes, this is not unusual. The GTM (which RBV is based on) is just a GLM where it uses a linear model to estimate the ROI intensities. There are no constraints on positivity. When the model is not exactly correct (which it never will be) it is possible to get negative values. Usually this only happens in small ROIs and/or low uptake regions. Why might the model not be perfect? It could be a number of things: 1. the MRI-PET registration is inaccurate 2. the MRI segmentation is inaccurate 3. the ideal segmentation is not reflected by the MRI segmentation (eg, one of the segmentations has a hot spot)

It could also be the case that the OSEM positivity constraint causes bias in low uptake regions (the GLM expects two-sided Gaussian noise)

On 8/28/20 7:27 PM, Albrecht, Daniel S. wrote:

Hello,

I’ve been running RBV partial volume correction on some amyloid PET data recently using mri_gtmpvc, and for ~90% of subjects the resulting rbv.nii.gz looks as expected with the exception of negative values in the bilateral accumbens (see attached image, thresholded from 0-4000 in scanner units). Any ideas why this might be happening?

This is the command I’ve been using:

mri_gtmpvc --i $PETimg \ --reg $identity_mtrx  \ --seg $FSdir/${subject}/mri/gtmseg.mgz \ --psf 6 --default-seg-merge --auto-mask PSF .01 --rbv \ --no-rescale --threads 8 \ --o $outname

Thanks very much,

Dan

---

Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer