PVS detection pipeline

AFNI Message Board
1

Hello all,

I am creating a perivascular spaces detection pipeline. The rough outline is

  1. Create FS anatomical masks of WM regions (mri_binarize/mri_convert)
  2. Erode masks to remove edges (3dmask_tool)
  3. Apply classification to full SurfVol (3dSeg)
  4. Compute overlap and isolate GM found within WM (3dcalc)
  5. Cluster those areas into proposed PVS (3dClusterize)

I find this is doing a decent job at identifying PVS, but definitely missing some. I am wondering if

  1. Increasing the image contrast (which function would that be?) would make it easier to identify some abnormalities in the WM
  2. Changing the parameters on 3dSeg could be helpful? I know the goal in WM classification would be to smooth out small abnormalities but I am specifically looking for those, so I want to discourage smoothing/grouping.

I'm looking for any help/advice, thanks in advance!
Leela Srinivasan

2

For contrast adjustments, 3dUnifize or 3dLocalUnifize would probably do. Anisotropic smoothing with 3danisosmooth could work for this kind of process.

3

Thank you so much! 3dUnifize with -GM worked well.

1 Like
4

Hi, Leela-

Just to mention that we recently did some work on a PVS project, which might have some useful aspects; please see here if interested. And I'm sure Gael would be happy to answer questions.

For tissue segmentation, if you have run FreeSurfer, have you looked at using those tissue masks? You can check whether those or 3dSeg's seem more accurate around areas of interest.

--pt