PVC will increase signal in some regions and will decrease it in others. Eg, for FDG it will increase it in GM and decrease it in WM. The reason you are seeing such a wide range is that the MG correction subtracts the WM then divides by the GM  partial volume fraction (PVF, a number between 0 and 1).  This requires a mask of some sort because there will be some voxels where the GM PVF is 0 (and you can't divide by 0). When you ran mri_gtmpvc with the --mgx option, you had to supply a threshold (eg, .01). This is the minimum allowed PVF before the voxel is masked out. If you used .01, then the multiplier could be as large as 1/.10 = 100, which agrees with the range you are seeing. Because of this problem, you must do the MG voxel-wise analysis on the surface where the GM PVF will be high. For subcortical analysis, you must use a mask of subcortical GM structures. The mask must be used when smoothing because you must only smooth within the mask (not such a problem on the surface). Note that there are many paper that use MG in a volume-based voxel-wise analysis; in my opinion, these studies are invalid.

Hi Dr Greve, Thank you very much for the great answers.

Kindly, I have one last question. 

The range of the signal intensity within the voxels in the original SUV maps is min=0.00 and max=3.017629

For the mgx images it is min=-43.74384 and max=88.468712

The difference in the range of signal intensity in the mgx images is largely wide. It seems that PVC increases the signal intensity. Is this correct?

Am I doing something wrong. I plan to include these images in voxel-wise analysis so I am curious about this difference between the images.

Thanks for any clarification!

 

On 8/18/17 10:20 AM, John Anderson wrote:

Hi Dr Greve,

I followed the steps in WIKI to do SUV-surface based analyses + PVC. I have the following questions and I highly appreciate your input:

 

1. Why the dimension of the images (mgx.gm, mgx.ctx.gm and  mgx.ctx.subgm) is not like the original SUV image that has been fed to the pipeline. i.e. I start with image-dimensions 128X128X128 : 2X2X2 and end up with 79X113X102 : 2X2X2

 

Also FOV is different as  well between the original SUV image  /256/ and the output mgx images /158/. How this happen? I am I doing something wrong?

I set up mri_gtmpvc to reduce the field of view to a bounding box around the head to reduce memory and computational loads. You can turn this off with --no-reduce-fov

 

2. Some voxels in the mgx images has negative signal intensity. is this normal?

Yes. MG works by estimating the contribution of non-GM to GM and subtracting it out. If the estimate is too high, then it can cause negative values.

 

Thank you for any clarification.

John

 

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