[caret-users] Creating a surface-based atlas from a volumetric-atlas

HINDRIKS, RIKKERT rikkert.hindriks at upf.edu
Tue Nov 4 05:23:26 CST 2014


Hi Donna,

To get a feeling for the registration process in Caret, I start with
performing a spherical registration of a human surface
to the PALS-atlas. I have extracted the surface and generated a border
projection file containing the required cuts and
landmarks. However, when I want to perform the registration, I get a
massage saying that Caret cannot find the target
border projection file. I used this file:

http://sumsdb.wustl.edu/sums/archivelist.do?archive_id=6057499

and indeed, it seems that there is no such file (nor coordinate files for
the fiducial and inflated surfaces). Are some
files missing or do I do something wrong?

Thanks and kind regards,
Rikkert




On Fri, Oct 31, 2014 at 4:27 PM, Donna Dierker <donna at brainvis.wustl.edu>
wrote:

> On Oct 30, 2014, at 11:29 AM, "HINDRIKS, RIKKERT" <
> rikkert.hindriks at upf.edu> wrote:
>
> > Dear Donna,
> >
> > Thanks for your fast response, I appreciate it!
> >
> > My situation is as follows:
> >
> > On the one hand, I have a group-averaged T1-weighted image, together
> with a volumetric atlas (that is, an integer labeling of the
> > voxels) as well as a structural connectivity matrix (obtained via
> fiber-tracking on the group-averaged diffusion-weighted image). On
> > the other hand, I have a T1-weighted image of an individual monkey. My
> aim is to obtain a surface atlas (derived from volumetric atlas)
> > for the individual monkey.
>
> This is an interesting scenario, and I've not encountered it before.
>
> > Could I first to a volumetric-registration of the individual image to
> the group-averaged image and subsequently project the induced
> > labeling of the voxels of the individual image to the individual surface?
>
> This seems reasonable and not too hard.  The lower variability in macaque
> folding may make it less problematic than for humans.
>
> > Or do I have to extract the surface of the group-averaged
> > image, project the volumetric atlas to it, and subsequently perform a
> spherical registration of the individual surface to the group-
> > averaged surface?
>
> People do extract surfaces from group averaged anatomical volumes for some
> purposes, but I doubt it will be worth it in this case.  I hope others will
> voice their opinions if they feel otherwise.
>
> > The first approach seems more straightforward, but I don't know if it is
> correct. Also, a complication with the second approach is that
> > the extracted surface from the group-averaged image looks worse than
> that extracted from the individual image (it is entirely ok, except
> > for that the primary visual cortex has a large part missing at the
> medial side).
>
> This is to be expected.  A more reasonable thing to do if you want an
> average surface is generate surfaces for the individuals and compute an
> average from them.  You probably don't have those surfaces, so honestly I'd
> try the first option and vet the resulting mapping using the
> T1+contour+volumetric-overlay view.
>
> Still another option would be to use surface based registration to get
> your individual monkey in register with the F6 atlas (part 3,
> http://brainvis.wustl.edu/wiki_linked_files/documentation/Caret_Tutorial_Sep22.pdf)
> or Donald McLaren's population average macaque atlas (
> http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2659879).  You could do
> something like this:
>
> * volumetrically warp your atlas goodies to match the mean anatomical
> McLaren image.
> * surface-based register your individual macaque atlas to the McLaren
> standard mesh surface.
> * map your warped atlas goodies to the McLaren population surface.
> * view your mapped results on your individual's standard mesh surface.
>
> But that second step isn't trivial, and your easier route might suffice.
> So I'd give that a go first.
>
> > And Donna, could you please tell me how to create a paint file from a
> nifty-file? (the atlas I have is saved as a nifti-file)
>
> In caret5, Attributes: Map Volume to Surface and choose paint.  But
> getting the color lookup is a bit messy.  The newer CIFTI format contains a
> label lookup table, and we also have a nifti extension for that, but it's
> not understood by other viewers.
>
> Caret5 has two volume formats that store that color lookup info:  AFNI,
> using a caret-specific header extension, and wustl's IFH/4dfp.  The ifh
> header lists the lookup, with an offset of 2.  (Like I said:  Messy.)  I
> can point you to some helpful threads on caret-users if needed.
>
> > The background is that we want to construct a computational model of
> cortical dynamics using the structural connectivity information.
> > An alternative, I guess, would be to spherically register the individual
> surface to the F99 template and subsequently, use the CoCoMAc
> > or other available connectivity data. The drawback of this, however, is
> that the strength of connections is more or less qualitative, hence
> > not so well-suited for modeling. If you think, though, that this is the
> best option for the creation of a surface-atlas, then I will go for it.
>
> See how far you get with the easy option.  If not far enough, then it's
> possible the McLaren atlas mesh is in register with F99.  I can't recall
> clearly, but read that link above for more info.
>
> > Thanks a lot Donna, and kind regards,
> > Rikkert
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > On Thu, Oct 30, 2014 at 3:13 PM, Donna Dierker <donna at brainvis.wustl.edu>
> wrote:
> > On Oct 29, 2014, at 10:56 AM, "HINDRIKS, RIKKERT" <
> rikkert.hindriks at upf.edu> wrote:
> >
> > >
> > > Dear all,
> > >
> > > I have an averaged T1-image and co-registered volumetric atlas of the
> macaque brain (which has been digitized by a collaborator) and want to
> derive from it a surface-based
> > > atlas. Subsequently, I would like to use this atlas to get a
> parcellation of the cortical surface of an individual macaque brain). How
> should I approach this problem?
> > >
> > > I have extracted the cortical surface from the averaged T1-weigthed
> scan.  Should I now
> > > just label each cortical vertex by determining to which ROI it
> belongs? And what if some vertices fall outside all ROI's? Also, the result
> does not look so smooth as existing atlases.
> >
> > It sounds like you need to map the volume(s) onto the surface.  It also
> sounds like these are discrete parcellations (ROI/label/paint) as opposed
> to probabilistic atlases, since it sounds like it is an individual monkey's
> data, rather than group data.  It would be helpful to clarify this.
> >
> > Assuming it is ROI/label (i.e., each intensity value -- e.g., 1, 2, 3, …
> -- corresponds to a region -- e.g., cingulate, arcuate, …), then I would
> map it as a paint volume.  I believe doing so constrains the mapping
> algorithms, but I am not certain.
> >
> > If you load your anatomical T1 with your surfaces and toggle on the
> surface contours (Volume Surface Outline, on the D/C page selection), then
> you can overlay the volumetric atlas over these two anatomical underlays
> (T1+surface contours) to look for regions where the surface does not
> intersect the atlas.  I see three choices:
> >
> > * fix the volumetric atlas data
> > * fix the surfaces, so the intersection is improved
> > * accept the fact that there are real holes in your data
> >
> > You will be better equipped to make that choice when you are looking at
> T1+surface contours+volumetric-atlas.
> >
> > > And to parcellate an individual macaque brain, can I register both the
> surfaces (that is, the template surface and the individual surface)
> spherically?
> >
> > Registering an individual monkey brain to a monkey atlas (e.g., F99)
> isn't really parcellating it, but there are parcellations already on the
> F99 atlas, so if you use spherical registration to register your monkey to
> F99, then you could look at the F99 parcellations overlaid on your monkey's
> surface.  But it's not a quick or easy process.  You need to draw
> registration borders.  (Though there are other registration algorithms out
> there that use sulcal maps and/or other data to automatically derive the
> deformation.  I encourage others to chime in if they ones they have used
> and found not too hard.)  How would you be using the registered surface?
> >
> > (Sorry for the delayed reply, but it wasn't a quick one. ;-)
> >
> > > Thanks a lot,
> > > Rikkert
> > >
> > >
> > > _______________________________________________
> > > caret-users mailing list
> > > caret-users at brainvis.wustl.edu
> > > http://brainvis.wustl.edu/mailman/listinfo/caret-users
> >
> >
> > _______________________________________________
> > caret-users mailing list
> > caret-users at brainvis.wustl.edu
> > http://brainvis.wustl.edu/mailman/listinfo/caret-users
> >
> > _______________________________________________
> > caret-users mailing list
> > caret-users at brainvis.wustl.edu
> > http://brainvis.wustl.edu/mailman/listinfo/caret-users
>
>
> _______________________________________________
> caret-users mailing list
> caret-users at brainvis.wustl.edu
> http://brainvis.wustl.edu/mailman/listinfo/caret-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://brainvis.wustl.edu/pipermail/caret-users/attachments/20141104/8d378c3f/attachment.html>


More information about the caret-users mailing list