External Email - Use Caution
Hi Krzysztof,
Samseg will "eat" anything you throw at it. That doesn't mean that the type
of input data doesn't matter: the higher the contrast-to-noise ratio for
the specific structure(s) you're interested in, the better the results will
be. Note
External Email - Use Caution
Hi Stefano,
For samseg isn't there also an option to disable the VAE shape model? I
forgot the details, but if false positives are not an issue maybe that
could also help?
Koen
On Friday, January 5, 2024, Stefano Cerri wrote:
> External
External Email - Use Caution
Hi Reza,
The lesion class in samseg is a bit special, as the internal lesion
probabilities are thresholded (using the value passed on via the
--threshold flag, or a default value 0.3 if you didn't use that flag) to
obtain the voxels labeled as 99 in
External Email - Use Caution
Hi Alberto,
As you already indicated, it's really a combination of an unsupervised
clustering method with a supervised shape model, so the dichotomy between
the two doesn't neatly apply. When I think about this type of method, I
consider it an
@DESKTOP-TILLA3B:/usr/local/freesurfer/7-dev/subjects$
>
>
> The image that I am trying to test with this process is a standard
> T1-weighted image from ADNI, and Freesurfer had been able to run the basic
> segmentation (-recon-all, took about four hours and ran to completion
&g
External Email - Use Caution
It looks like you have an extra slash in front of the output file name, so
you're trying to write in the root directory...
On Tuesday, March 8, 2022, wrote:
> External Email - Use Caution
>
> Dear experts,
>
> Having installed Freesurfer on
External Email - Use Caution
Hi Peder,
This is tricky -- although samseg is intended to adapt to differences in
contrasts and/or number of contrasts, it will often not put the
segmentation boundaries at the exact same locations depending on the input
type. Such inconsistencies
External Email - Use Caution
Hi Randolph,
When you visualize the segmentation produced by samseg on top of its input
image, does the segmentation look correct? If so, the problem is very
likely not the affine registration code. (If the segmentation looks bad,
please try samseg in
External Email - Use Caution
Hi Rana,
The samseg tool in FreeSurfer will segment some non-brain tissues. If
that's not sufficient for your needs, you can also check out a more
specialized version of samseg distributed with the new/upcoming version of
SimNIBS -- cf.
External Email - Use Caution
Hi Simon,
This could be an issue with the affine registration or the subsequent head
masking step in samseg, both of which have been upgraded/improved in the
development version.
Would you be able to try the dev version and see if this resolves the
External Email - Use Caution
Hi Alberto,
In the current implementation the volume reported in the table is simply
the sum of the lesion posterior probabilities, i.e., it is not affected by
the threshold used to create the crisp segmentation result for the lesions.
A few months
External Email - Use Caution
Hi Andrew,
No, I don't think we have seen something like that before.
I'm a bit doubtful about the "enhanced T1 images", though - for sure there
is more information for samseg to work on when you feed it both T1 and
FLAIR, compared to some fused
Van Leemput k...@nmr.mgh.harvard.edu:
Hi Joana,
This looks like a visualization issue rather than a segmentation error to
me. How did you visualize this?
Koen
On Jun 13, 2014 2:59 PM, Joana Braga Pereira jbragapere...@gmail.com
wrote:
Dear all,
I'm running the GEMS tool in Freesurfer
Hi Joana,
This looks like a visualization issue rather than a segmentation error to
me. How did you visualize this?
Koen
On Jun 13, 2014 2:59 PM, Joana Braga Pereira jbragapere...@gmail.com
wrote:
Dear all,
I'm running the GEMS tool in Freesurfer 5.3 to measure the volumes of
hippocampal
Hi Mike,
We have found that the volumes obtained by summing the subfield
volumes discriminates a bit better between controls and AD subjects
than the standard FS hippocampal volumes.
Hope this helps,
Koen
On Tue, Apr 15, 2014 at 7:11 PM, Mike Strupp-Levitsky
ms4...@columbia.edu wrote:
To
Hi Salil,
You're right: the conversion factor to mm^3 is 1/8, since every voxel is
0.5mm isotropic.
The partial volume modeling code is there but not thoroughly tested. Using
it is not recommended.
Hope this helps,
Koen
On Dec 22, 2013 2:10 AM, Salil Soman salso...@stanford.edu wrote:
Hi,
Hi,
I believe recon-all -s NC -hippo-subfields should do that.
Please let us know if it doesn't.
Koen
On Sep 19, 2013 9:00 AM, Anupa AV av.an...@yahoo.com wrote:
Dear FS experts,
I've a query regarding the recon-all analysis.
I ran recon-all using the following command
eg.
recon-all -i
Hi Stefano,
Can you do ldd kvlApplyTransform and send us the result?
Koen
On 3/22/13, std...@virgilio.it std...@virgilio.it Hippocampal Subfields
processing Fri Mar 22 16:56:31 CET 2013\n mkdir -p
/Applications/freesurfer/subjects/subject_prova/Diff02/Diff02/tmp/subfieldResults\n
Hi Lorenzo,
The correct way to calculate the expected (not an integer value) number of
voxels belonging to a subfield is by summing ( intensity / 255 ) across all
voxels. Not sure where things go wrong: the integer values you report are a
bit suspicious, or perhaps some rescaling of image
Hi Michael,
I can answer the subfield part - perhaps someone else (Bruce?) can
chime in regarding aseg.
For the subfields, there isn't really a standard routine to go in and
alter the segmentation results. You could try to alter the aseg
hippocampus segmentation as this is what is used to
Gonzalez Escamilla
ggon...@upo.es wrote:
Dear Fresurfer experts and Koen Van Leemput
I have made the hippocampal subfield segmentation of a group a subject in
which is supposed to take the hippocampus from the aseg segmentation and
calculate the new subfields segmentations inside this hippocampus
Hi Gabriel,
1. Using pre-processed data obtained with an older version of
FreeSurfer should be OK, as long as you remember that the obtained
results might/will be different from the ones you'd get if you had run
the whole analysis from scratch.
2. The generic label hippocampus is used to denote
Hi Lorenzo,
The hippocampal subfield segmentation routines are only available from
FreeSurfer 5.1, so you'll need that version to run the code.
If all your data is already pre-processed with an older version of
FreeSurfer, you might be able to simply run the hippocampal subfield
segmentation
Hi Irwin,
The command should do exactly as you describe, but it seems to work
when I try it.
I did notice that the filename you specified
(hippoposterior_right_CA4-DG.mgz.gz) has an extra hippo in front -
could you please double-check?
Thanks,
Koen
On Sat, Apr 7, 2012 at 10:07 PM, Irwin,
,
No worries.
The result of grep is:
500 right_CA2_3 17 85 136 0
550 left_CA2_3 17 85 137 0
Ed
On 4 Apr 2012, at 12:41, Koen Van Leemput wrote:
Hi Ed,
Can you please do
grep _CA2 $FREESURFER_HOME/FreeSurferColorLUT.txt
and let me know what the result
-Darwin-leopard-i686-stable-pub-v51.0..dmg,
date 26/05/2011.
Ed
On 4 Apr 2012, at 13:09, Koen Van Leemput wrote:
OK, I see why this is not working. We've been changing some name
conventions in our internal FreeSurfer repository after the public
release of version 5.1, and somehow you managed
Hi Ed,
What version of the FreeSurfer build are you using?
Also, could you please let me know what the result is of ls
posterior_left_CA2*?
Thanks,
Koen
On Tue, Mar 27, 2012 at 8:29 AM, Ed Gronenschild
ed.gronensch...@np.unimaas.nl wrote:
Hi,
I followed the instructions to visualize the
Hi Joshua,
The correct way to obtain binarized maps would be to visit each
voxel in turn, compute the probability that it belongs to each of the
subfieds, and pick the subfield with the highest such probability.
Having said that, it should not be the case that there are two or more
subfields
Hi Rob,
It seems the internal 5.1 version of FreeSurfer you're running was
somehow compiled on 12 April 2011 (?), whereas the official release
date was 24 May 2011.
Could you please double-check if this is indeed the case? If it is,
could you please run the official release instead? We changed a
Hi Oliver,
You're right: internally we're upsampling the images to 0.5mm
isotropic resolution, so if you had acquired 0.5mm isotropic images
from the start, that would actually be much better (provided you don't
simply trade off SNR for increased image matrix size). The catch with
this is that
Hi Rob,
I think this might be related to the same 5.1 version mismatch I
indicated in my previous email to you. Could you please try if things
start working when you use the official FS release?
Thanks,
Koen
On Mon, Feb 6, 2012 at 1:59 PM, Rob Tennyson te...@nmr.mgh.harvard.edu wrote:
Hi
Hi Rob,
Same reply - the current scripts should do this automatically for you...
Please let me know if you run into trouble with this,
Koen
On Tue, Feb 7, 2012 at 11:18 AM, Rob Tennyson te...@nmr.mgh.harvard.edu wrote:
Hi Experts,
Is there a way to add the CA-4/DG subfield to the output
Hi Joana,
The smallness of our CA1 compared to other protocols seems to be
caused by disagreements in the fields of where the subfield boundaries
should be located. I've talked to several experts on this and they all
are confident that their own definition is the only right one. :-)
As for the
Hi Tetiana,
Could you please do
grep left_CA1 $FREESURFER_HOME/FreeSurferColorLUT.txt
and see what the result is? There have been some naming convention
changes (left_CA1 - left-CA1, etc), so perhaps this is the problem?
Thanks,
Koen
On Thu, Dec 8, 2011 at 7:48 AM, Tetiana Dadakova
Hi Allan and Joshua,
The hippocampal subfield segmentation really just uses a few output
files generated by the usual FreeSurfer volumetric stream, so the
short answer is that it should be perfectly possible to use 5.1 to do
hippocampal subfield segmentation on FreeSurfer results processed with
The answer is close to what Shashwath suggested, except that the
Hippocampus segmentation is really computed at the same time as all
the other subfields. For unfortunate reasons, the tail of the
hippocampus that is not further subdivided in our manual training
segmentations was called Hippocampus
, 2011 at 6:14 AM, Koen Van Leemput
k...@nmr.mgh.harvard.edu wrote:
The answer is close to what Shashwath suggested, except that the
Hippocampus segmentation is really computed at the same time as all
the other subfields. For unfortunate reasons, the tail of the
hippocampus that is not further
Hi Joshua,
The version of the hippocampal subfield segmentation routines
currently in FreeSurfer will only work out of the box on the usual
FreeSurfer-resampled 1mm isotropic data. If you are technically
skilled, it should be possible to tweak things to make it work on
other data (we don't assume
Hi Mohan,
I'm afraid the smaller structures you mentioned aren't generated by
our automated segmentation tools. As for the Left-Hippocampus and
Right-Hippocampus labels, these are parts of the tail of the
hippocampus where the people doing the manual segmentations the
automated method was trained
Hi Shashwath,
The label hippocampus was used in the tail of the hippocampus, where
the manual labelers (who labeled the data our automated method is
trained on) stopped trying to distinguish between the different
subfields. If you want to compute the volume of the whole hippocampus,
then you
] Segmentation of hippocampal subfilelds
To: Koen Van Leemput k...@nmr.mgh.harvard.edu
Hello,
I would like to thank you for the information, but the problem is, I am a CS
student working on this and I dont have much idea about it. Basically my
task is to convert those images into ROI mask
Hi Mohan,
The segmentation results are given as probability maps: a (unsigned
8-bit) value of 0 means probability 0 , 255 means probability 1, and
everything in between scales linearly.
The proper way to obtain crisp label maps from these images is to
assign each voxel to the label map that has
overlapping?
Or maybe transfer my own image into this 0.5x0.5x0.5 mm^3 space. How can I
achieve this?
Thanks,
Wayne
On 06/07/11 5:59 AM, Koen Van Leemput k...@nmr.mgh.harvard.edu wrote:
Hi Tricia,
Yes, that summarizes it neatly. :-)
Koen
On Mon, Jul 4, 2011 at 10:44 PM, Tricia Merkley
Hi Tricia,
Yes, that summarizes it neatly. :-)
Koen
On Mon, Jul 4, 2011 at 10:44 PM, Tricia Merkley tmerkl...@yahoo.com wrote:
Hi,
I understand that the models for the hippocampal subfield segmentation were
created using high resolution MRI scans. But it sounds like this procedure
can
Hi Juergen,
This looks like an error very early in the hippocampal subfield
segmentation routines. Could you please send me the file
hippo-subfields.log (found in the scripts subdirectory of the
subject in question), and ideally the files nu.mgz, aseg.mgz, and
transforms/talairach.xfm files
Hi New Fei,
The subfield volume estimates are given in number of voxels of size
0.5mm isotropic - you'd need to divide by a factor of 8 to obtain
measurments in mm^3.
The sum of the subfield volumes will not be the same as the total
hippocampal volume in aseg.stats because the used
:
Hi Victor
we treat it as any other label - it's manually labeled in the training
subjects and propagated into the aseg. Koen Van Leemput (ccd) has some
tools for segmenting them from multi-spectral data that probably works
better though (they aren't that well defined only on a T1
We have recently developed a method for automatically segmenting the
hippocampal subfields from ultra-high resolution MRI data. The software code
is not yet in FreeSurfer, but will be in the future. It produces volumetric
segmentations, rather than surface representations, though.
- Koen
48 matches
Mail list logo