On 1/29/19 11:21 AM, Adam Martersteck wrote:
External Email - Use Caution
Hi FreeSurfer team, I have two comments/questions about PET PVC & Doug’s 2016 Neuroimage paper:
- It looks like the 2 rigid transforms created during mri_gtmpvc
found at [$pvc_directory/aux/anat2bbpet.lta] and [$pvc_directory/aux/bbpet2anat.lta] are misnamed and the file names should be swapped. I noticed this isn’t a problem during mri_vol2surf (e.g. the PetSurfer wiki) because mri_vol2surf is “smart” and will automatically detect and invert the transform -- but it was causing problems for me when I used them during mri_vol2vol thinking the bbpet2anat.lta was from pet space to anat space. I apologize if you already knew about this – I couldn’t find anything in the FSv6 “Known Issues” or on the PetSurfer page.
The names are actually correct (if you look inside the ltas you will see). The problem is that mri_vol2vol expects the registration to be anat2pet (for reasons going back 20 years). I'll add something to mri_vol2vol to make it smart as well.
- I am using the MG PVC stream and a fairly conservative GM PVF
threshold of 30% (similar to Doug’s paper). I noticed when the PVF is < 0.3 it sets the corrected [mg.nii.gz] image to zero in those voxels. Intuitively, I would have thought this would just “cap” all values < 0.3 to 0.3 and divide voxels by 0.3 to get a MG corrected voxel. I went back to Doug’s paper and found, indeed, the paper says, /“In this study, we varied the threshold from 1% to 50%. The TACs in voxels that did not meet this criterion were set to zero.”/
So when I take the mg.nii.gz and use vol2surf as Doug did, I get a significant number of “patches” of zero values across the cortex (see attached image; every voxel along the normal is zero). I would then have smoothed this surface -- artificially lowering the entire neighborhood of vertices. Did you guys do something else for the Neuroimage paper when you were going from vol2surf? Or did you smooth the zeroes into the surrounding vertices? Maybe this isn't a problem in healthy brains and the projfrac 0.5 voxel is always a non-zero?
We just smoothed the 0s in. Note that we do say "This justification for 0.3 cannot be generalized to other studies with different age groups and PSFs." But we also say that the results did not change much at lower thresholds. I would just reduce the threshold. I usually use .01 for the surface-based analyses because we are pretty confident that we are sampling in the middle of cortex.
Since I am working with significantly atrophied Alzheimer brains -- would “capping” the PVF make sense here to balance over correcting noise while avoiding non-zero vertices? Is there a command in fscalc to set all values between 0.0001 and 0.3 to 0.3 so I could do this?
I think that would probably produce some strange stats.
Thanks! Adam
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer