Hi Bruce,
Thanks. So it sounds like FreeSurfer 4.5 should now be adequately filling ventricles in the general case, and that -bigventricles offers didn't reach a level of robustness you were happy with, so best to avoid that? What about the flag -fillven on mri_edit_wm_with_aseg? Any use?
In terms of the fix mentioned below, the fix literally is:
1. recon-all -all 2. mri_binarize --i $SUBJECTS_DIR/$PD/mri/aparc+aseg.mgz --match 4 --match 43 --match 5 --match 44 --o $SUBJECTS_DIR/$PD/mri/vents.mgz --binval 255 3. Add vents.mgz to wm.mgz 4. recon-all -autorecon2-wm
so, it could well be affecting the results more by virtue of the fact that we essentially perform a re-run?
Thanks ever so much.
Matt
----- Original Message ----- From: "Bruce Fischl" fischl@nmr.mgh.harvard.edu To: "Matt Clarkson" clarkson@drc.ion.ucl.ac.uk Cc: freesurfer@nmr.mgh.harvard.edu Sent: Friday, November 13, 2009 2:44:03 PM GMT +00:00 GMT Britain, Ireland, Portugal Subject: Re: [Freesurfer] Filling Big Ventricles
Hi Matt,
yes, the filling procedure has evolved and is better in current versions than previous ones.
cheers, Bruce
On Fri, 13 Nov 2009, Matt Clarkson wrote:
Hi Bruce,
I agree with your point, however this method was instigated by a previous person at our institution, so I'm looking at it afresh. Also, the addition of regions 4, 43, 5 and 44 does do "something". I agree though that when I look at wm.mgz using tkmedit on a vanilla FreeSurfer run, it appears that the white matter is a grey colour, and that the vents have been filled with white (colour 255).
So, is it possible that a previous version of FreeSurfer did not by default try and fill the vents in, and now version 4.5 does by default attempt to fill the vents in?
Cheers
Matt
----- Original Message ----- From: "Bruce Fischl" fischl@nmr.mgh.harvard.edu To: "Matt Clarkson" clarkson@drc.ion.ucl.ac.uk Cc: freesurfer@nmr.mgh.harvard.edu Sent: Friday, November 13, 2009 2:25:16 PM GMT +00:00 GMT Britain, Ireland, Portugal Subject: Re: [Freesurfer] Filling Big Ventricles
Hi Matt,
the -bigventricles stuff is something I messed with for a while, but didn't get to work robustly and haven't gotten back to it. I'm not sure I understand your fix though. If the aseg properly labels the ventricles, then they should be filled in the wm.mgz, and if the aseg doesn't, then your fix doesn't help. Do you manually correct before doing this?
cheers, Bruce
On Fri, 13 Nov 2009, Matt Clarkson wrote:
Hi there,
we have various Alzheimer cases, where the FreeSurfer pipeline can occasionally struggle to correctly segment the ventricles, which results in white matter and pial surfaces "leaking" into the ventricles.
I have tried using -bigventricles passed into the autorecon2 stage, but I can't tell if its doing much. I checked the log files, and you can see -bigventricles passed to several binaries such as mri_ca_register and mri_ca_label, so the flag is definitely being passed from the command line.
Is the -bigventricles meant to be used in this way, and what should it be doing?
Also, I notice that the binary mri_edit_wm_with_aseg has a flag -fillven, but I can't find the documentation to say what it does and if I should be using it.
So, we have previously used a technique whereby we run freesurfer once then create a vents image like: mri_binarize --i aparc+aseg.mgz --match 4 --match 43 --match 5 --match 44 --o vents.mgz --binval 255
and then add this into the wm.mgz to fill some holes in. I have attached two .png files to illustrate the effect. In this example, I have chosen the slice where the surfaces just start to leak into the ventricles. So, on these snapshots, the problem looks quite small, but on other subsequent slices, you can get fairly large leakage into the ventricles.
So, with these options, what would be the best way to cope with very abnormal ventricles?
Many thanks in advance.
Matt Clarkson
Senior Research Associate Dementia Research Centre UCL Institute Of Neurology Queen Square London WC1N 3BG
m.clarkson@ucl.ac.uk Tel: 08451 555 000 ext. 723653 Fax: 020 7676 2066