Hi Sebastien
that isn't how the algorithm works. If I had to guess, I would guess that
what you are seeing is locations with high through-plane curvature so even
though a voxel looks like it is all CSF in fact the surface cuts through
it. You can see this by moving back and forth a few slices. If you upload
the dataset I'll take a look, but I've run those functions on hundreds of
datasets and am relatively confident they are correct (although of course
bugs happen!)
cheers
Bruce
On Fri, 11 Sep 2015, Sebastien Proulx wrote:
Hello Bruce,
I ran some tests overnight. By cranking up the memory of my virtual machine to
its maximum (close to the 16GB of my actual machine),
mri_compute_volume_fractions could run through at resolutions of -r 0.333 and
-r 0.25 :-)
FYI, it took 13min at the default resolution of 0.5, 50min at 0.333 and 150min
at 0.25, with my Intel Core i7-3770 3.4GHz processor.
More importantly, at both 0.333 and 0.25, the sulcus that did get filled at the
default resolution of 0.5 did get filled. But I found other smaller sulci that
did not get filled even at 0.25... See RAS coordinate 19 -44 -34 in the csf
images at
https://www.dropbox.com/sh/tw534pw3z4v9xuh/AACth_3UesAWyRr1i5aM4xd0a?dl=0 with
self-explanatory names for one example, and the attached .png images for a
quick look.
I will try even higher resolutions later. I can live with some procedure where
I just visually make sure that this problem is not happening in my MRS voxels
and do some manual correction to minimize the damage. But I wish the function
was more robust.
I guess that the function is filling wm from the inside-out and csf from the
outside-in, and compute gm from whatever is not wm or csf, right? If it is the
case, I also guess that when entering a sulcus, csf filling stops when the
space between the pial surfaces defining that sulcus is too small relative to
the resolution used. I further guess that this happens less, or not at all, for
wm since the wm surfaces rarely get close to each other (contrarily to the pial
surface). And I finally guess that filling wm volume from the inside-out and a
wm+gm volume also from the inside-out would allow to compute all three tissue
fractions appropriately, while avoiding the problem of tight sulci we observe.
Is this series of guess correct? Do you think fooling the function by just
changing names of the wm surface to pial surface would allow me to test this
idea?
Thanks a lot for your precious help!
Sébastien
-----Original Message-----
From: Bruce Fischl [mailto:fis...@nmr.mgh.harvard.edu]
Sent: Thursday, September 10, 2015 10:28 AM
To: Sebastien Proulx
Cc: freesurfer@nmr.mgh.harvard.edu
Subject: RE: [Freesurfer] mri_compute_volume_fractions not filling some sulci
properly
there aren't really any workarounds other than using a machine with more ram.
Otherwise you can reduce the resolution. Maybe try -r 0.3333
Bruce
On Thu, 10
Sep 2015, Sebastien Proulx wrote:
ho yes, I meant 11.9GB. I could try to crank it up (or closer) to the physical
limit of my machine at 16GB. Would that be enough? Are there work around, since
going higher than 11.9GB might cause problem (memory swapping) between my
virtual machine and my native os.
-----Original Message-----
From: Bruce Fischl [mailto:fis...@nmr.mgh.harvard.edu]
Sent: Thursday, September 10, 2015 10:15 AM
To: Sebastien Proulx
Cc: freesurfer@nmr.mgh.harvard.edu
Subject: RE: [Freesurfer] mri_compute_volume_fractions not filling
some sulci properly
presumably you mean GB, not MB? That's not enough - a single distance
transform of a 256^3 volumes at 0.25 mm is 1GB of 4-8 byte voxels and
we have to hold a few in memory
On Thu, 10 Sep 2015, Sebastien Proulx wrote:
11.9 MB on a virtual Ubuntu 14.04.3 with VMware Player.
-----Original Message-----
From: Bruce Fischl [mailto:fis...@nmr.mgh.harvard.edu]
Sent: Thursday, September 10, 2015 10:06 AM
To: Sebastien Proulx
Subject: Re: [Freesurfer] mri_compute_volume_fractions not filling
some sulci properly
how much ram do you have in the machine? You probably need more. Also, can you
please cc the list on this type of email so others can answer and learn?
thanks
Bruce
On Thu, 10 Sep 2015, Sebastien Proulx wrote:
Hello Bruce !
Thanks for the tip. I tried it but it did not run through, it was killed
without much error message:
--------------------------------------------------------------------
-
-
-------------------------------
-------------------------------------------------------
bass@ubuntu:/mnt/hgfs/workOnDropbox/projects/150422_decodingHRF/HRph
e n omenology$ mri_compute_volume_fractions -r 0.25
$curSubj"_registerIdentity.dat" $curSubj_FS/mri/brain.mgz
$outVol"_PLUS_r"
setting resolution = 0.250
arg = 4
reading registration file
ha/ha_run3_tissueFractionMasks__tmp/ha_registerIdentity.dat
reading surface
/mnt/hgfs/workOnDropbox/projects/150422_decodingHRF/HRphenomenology/
s
2
5_HA_freesurfer/surf/lh.white
reading surface
/mnt/hgfs/workOnDropbox/projects/150422_decodingHRF/HRphenomenology/
s
2
5_HA_freesurfer/surf/rh.white
reading surface
/mnt/hgfs/workOnDropbox/projects/150422_decodingHRF/HRphenomenology/
s
2
5_HA_freesurfer/surf/lh.pial
reading surface
/mnt/hgfs/workOnDropbox/projects/150422_decodingHRF/HRphenomenology/
s
2
5_HA_freesurfer/surf/rh.pial
reading volume
/mnt/hgfs/workOnDropbox/projects/150422_decodingHRF/HRphenomenology/
s
2
5_HA_freesurfer/mri/aseg.mgz
reading movable volume s25_HA_freesurfer/mri/brain.mgz
filling interior of lh pial surface...
filling interior of rh pial surface...
filling interior of lh white matter surface...
Killed
--------------------------------------------------------------------
-
-
-------------------------------
-------------------------------------------------------
Could it have exceeded memory? I’ll rerun it, but in the meanwhile,
any tip to get more information for debugging?
Also, here attached is a sample script, and you can get a compressed
version of the FS folder for one subject here
https://www.dropbox.com/s/s0h8flnxonskzbu/s25_HA_freesurfer.tar.gz?dl=0.
Thanks a lot again and
Have a very good day!
Sébastien
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine
at http://www.partners.org/complianceline . If the e-mail was sent to you in
error but does not contain patient information, please contact the sender and
properly dispose of the e-mail.
_______________________________________________
Freesurfer mailing list
Freesurfer@nmr.mgh.harvard.edu
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is
addressed. If you believe this e-mail was sent to you in error and the e-mail
contains patient information, please contact the Partners Compliance HelpLine at
http://www.partners.org/complianceline . If the e-mail was sent to you in error
but does not contain patient information, please contact the sender and properly
dispose of the e-mail.
