Hi Nick,
It appears mri_compute_seg_overlap only works with .mgz files and not .label files, which is how I saved the ROIs. I tried renaming the labels as .mgz files, but that did not work. Any suggestions? Thanks.
the utility for comparing segs is 'mri_compute_seg_overlap', included with freesurfer. it compares two arbitrary segs. we use it to compare the manual seg with the auto seg to get Jaccard and Dice coefficients.
On Mon, 2009-08-17 at 12:03 -0400, Bruce Fischl wrote:
if all you want is reliability on the same volume then you don't need any reconstructions. Just manually label them in something like freeview and directly compute the overlap (Nick has some tools for this if you don't
it's pretty straightforward)
cheers, Bruce On Mon, 17 Aug 2009 jentis@nmr.mgh.harvard.edu wrote:
Hi Bruce,
Essentially, for this ROI reliability problem we've asked you about,
we
want to use the high-res volumetric data to make a refined
segmentation
of the amygdala. We can't do this by using the 1mm data because of
its
visual integrity. But, maybe what we can do (as per your email) is
make
the labels in the .38 volume data, map that to the 1mm data and then compare the reliability of those labels? Basically, this is a similar project to Brad Dickerson's high-res manual hippocampus segmentation.
We
just need to confirm that we can trace it twice with a high amount of overlap and accuracy. Thanks so much for your help!
Best, Jon
Hi Jon,
I would use the 1mm data to create the surfaces then map them to the hires data and do your tracing that way. You can then use the
surfaces for
various things like inter-subject registration. I'd need more details
to
give better advice, but generating surfaces from highres data is
hard, and
usually not necessary.
cheers, Bruce
On Mon, 10 Aug 2009 jentis@nmr.mgh.harvard.edu wrote:
Hi Bruce,
Yes, we have the 1mm data. However, the project is designed to manually trace an ROI based on the increased visibility in the
high-res
scans. We need to do reliability to test our method of manual segmentation. Perhaps I am misunderstanding, but it seems like
using
the 1mm surface data would defeat our purpose. Any suggestions? Thanks again!
Best, Jon
can't you get 1mm data to go with your highres data? For many
reasons
the high resolution data is usually quite poor for generating
surface
models.
cheers, Bruce On Sun, 9 Aug 2009 jentis@nmr.mgh.harvard.edu wrote:
> Hi Freesurfer team, > I am trying to get intra-rater reliability statistics on an ROI > manually traced twice (blindly). The scans are anisotropic (.38 x
.38
> x .8 mm) and so I have not run them automatically through the > Freesurfer segmentation because they fail after autorecon1. I
tried
> to > use the following command but have had problems: > mris_anatomical_stats -a ${s}/label/${h}.aparc.annot -f > cortical_thickness_stats/${h}_${s}.txt $s $h > > It seems this command requires certain data from the autorecon
process
> such as wm.mgz and other surface information that I cannot get
because
> autorecon fails on these types of scans. > > I have only traced one hemisphere (the Left) and the ROIs are
saved as
> label files; > an example is: > /autofs/space/phrenic_007/users/jon/hires_amy/Round_2/HRDEV126/ > whole_amygdala.label > > Is there another way to get reliability (pixel overlap, total
number
> of > pixels, etc.) for two ROIs? Would you suggest another command or
is
> there > a way to bypass those requirements for the mri_anatomical_stats > command? > > Thanks! > _______________________________________________ > Freesurfer mailing list > Freesurfer@nmr.mgh.harvard.edu > https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer > > >
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer