Closed sulantha2006 closed 5 years ago
Hi,
It would make sense to PV-correct the data from each centre with a centre-specific PSF and then apply some smoothing. However, if the post-smoothing required is large, then it may make more sense to smooth all centres to a common resolution, but then I wouldn’t apply the PVC to the smoothed data.
RBV is intended for correcting multiple regions; usually multiple regions within a tissue class. Applying RBV with a three class segmentation would likely result in biases coming from the GTM stage. I would suggest increasing the number of regions if you wish to use RBV. In the situation you describe, assuming that WM was completely uniform and it was possible to accurately measure the WM mean value, MG would be better. Although, personally I would choose to use RBV with a more detailed segmentation.
If this answers your query, please feel free to close the issue.
Kind regards,
Ben
Quick question, How do you define a large smoothing kernel? Bigger than the psf of the scanner? The HRRT scanner has a PSF of 2.4mm, but applying a smoothing of < 2.4mm isn't ideal for a PET analysis. Typical smoothing kernel is about 6-8mm, What would be the best option at this point. ?
Say, in a single center study with HRRT, If I want to run a voxel-base stats, which steps if recommended.
Hi,
pvc_simulate
can be used to post-smooth the data if you want. Generally, I would take the approach in 1; PVC then smooth back. However, depending on your voxel size, you might not see much benefit on your HRRT data. You should just try it and see. Also, is the 2.4mm PSF estimated for a brain/brain-like object or point sources? If point sources, 2.4mm might be an underestimation of reconstructed resolution.
Hi. Thanks for the info.
I will get more details on how they estimated the 2.4 mm resolution.
What's the difference of pet_simulate vs 3d standard gaussian smoothing? I will read the doc but, just to know.
Thanks a lot in advance.
On Sat, Nov 10, 2018, 5:35 AM Ben Thomas <notifications@github.com wrote:
Hi,
pvc_simulate can be used to post-smooth the data if you want. Generally, I would take the approach in 1; PVC then smooth back. However, depending on your voxel size, you might not see much benefit on your HRRT data. You should just try it and see. Also, is the 2.4mm PSF estimated for a brain/brain-like object or point sources? If point sources, 2.4mm might be an underestimation of reconstructed resolution.
— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/UCL/PETPVC/issues/41#issuecomment-437574388, or mute the thread https://github.com/notifications/unsubscribe-auth/AA-V96sXAAoyZ_sd9gvs3-glK4K8tFoxks5utqvwgaJpZM4YPfbL .
pvc_simulate
performs 3D Gaussian smoothing. It is called this because it is was written as part of the test suite to simulate PVE.
I would like to know if there are drawbacks in applying a Gaussian smoothing kernel after the PVC correction to improve the SNR of the PET image.
We need to perform statistical analysis on PET images from multiple centers and the PET PSF isnt the same for the scanners. Also, one of the scanners is an HRRT scanner with a PSF of 2.4mm isotropic. We usually smooth the images to have all images from all centers in the same effective resolution (say 8mm) for the purpose of increasing the SNR for the statistical analysis.
My question is, Does it make sense to apply a smoothing after PVC,? I understand that this will remove the correction done by the PVC, I would like your expertise on it.
Also, If I do smooth to reach an effective resolution of 8mm, Can I use PVC with a kernel of 8mm to get an effective PVC?
Does it make sense to apply RBV with a 3 tissue probabilistic mask (GM, WM, CSF)? Would it be similar to a correction by MG?