[caret-users] caret spherical registration method
jidanz at hotmail.com
Wed Mar 3 23:55:13 CST 2010
Thanks for your answer. So in the manual you mentioned, they depict the registration process as two stages. Could you tell me more details about how do you measure the similarity between the landmarks and the spheres?
Thanks a lot.
Jidan--------------------------------------------------------------------------------------------------- Re: caret spherical registration method (Donna Dierker)
> Date: Mon, 18 Jan 2010 12:29:37 -0600 (CST)
> From: Donna Dierker <donna at brainvis.wustl.edu>
> Subject: Re: [caret-users] caret spherical registration method
> To: "Caret, SureFit, and SuMS software users"
> <caret-users at brainvis.wustl.edu>
> <63184.108.40.206.169.1263839377.squirrel at brainvis.wustl.edu>
> Content-Type: text/plain;charset=iso-8859-1
> Hi Jidan,
> First, today is a US holiday, so no one is in the office. I'm emailing
> from home.
> Second, you are not alone. Others have tried to do what you want to do,
> and had similar trouble. The good news is that David is working on a
> paper that describes the latest Caret registration algorithm. I don't
> know when it will be published -- probably this year.
> The bad news is that this is not the algorithm Caret has used since 2003.
> (And I don't know if the current Caret default algorithm is the new one or
> what has been used since 2003.)
> The one used 2003 until at least mid 2009 is described on page 27 of this
> Caret 4.6 tutorial:
> Here is the relevant excerpt:
> Outline of landmark-constrained spherical registration process.
> The landmarks used to constrain the deformation from one spherical map to
> another are derived from the individual and atlas border files specified
> in the Sphere Warp Parameters dialog. Caret first checks that the
> landmarks are matched in the individual and atlas border files (the
> identical number of contours, corresponding names in each file). If the
> designated files are based on borders from a flat map, they are first
> projected to the sphere. Landmark contours are then resampled (at a
> spacing specified in the Sphere Warp Parameters dialog) so that
> corresponding landmarks have identical numbers of points in the individual
> and atlas.
> A deformation sphere is generated for the individual and atlas by
> integrating the resampled landmark points into a regularly tessellated
> sphere that is selected from one of six available (74 nodes, 290 nodes,
> 1154 nodes 4610 nodes, 18,434 nodes, or 73,730 nodes, selected using the
> Sphere Resolution parameter). Usually, either the 4,610 or 18,434 node
> sphere is preferable. If, as in the tutorial, Deform Individual to Atlas
> is selected, the individual is identified as the source and the atlas is
> identified as the target . (If Deform Atlas to Individual is selected,
> the order is reversed.)
> The source deformation sphere is then deformed to the target deformation
> sphere by multiple cycles of a two-stage process that aims to preserve
> local shape characteristics of the source sphere while forcing the source
> landmarks into register with the target landmarks. In the first stage
> (landmark-constrained smoothing) the coordinates of all landmark points in
> the source sphere are repositioned to match those of the target sphere,
> and the surface is smoothed by an iterative process that keeps the
> landmark points tied to their target location while progressively drawing
> intervening points into register with the repositioned landmark points.
> The amount of smoothing can be adjusted using the Smoothing Parameters in
> the Sphere Warp Parameters dialog. The process can be viewed as it occurs
> by viewing the AUX configuration, with Nodes on and Tiles and Edges off.
> In the second stage, the deformed source sphere is subjected to spherical
> morphing, using the same process as in reducing distortions on the
> spherical map (Tutorial 5). In this case, the reference surface is the
> initial source deformation sphere, and the forces applied to each surface
> node work to reduce local distortions and further restore local shape
> characteristics. The distortions between the fiducial surface and the
> spherical surface are compensated for to some degree by the Correct for
> Spherical Distortion Relative to Fiducial parameter, whose default level
> is set to 0.5. A higher value (max = 1.0) can lead to improved
> compensation, but may also lead to instability and a poor result if pushed
> too far.
> The intermediate files generated during spherical registration can be
> viewed using the appropriate specification file
> (ia_target_with_landmarks.spec or ai_target_with_landmarks.spec, where
> "ia" signifies individual-to-atlas and "ai" signifies atlas-to-individual.
> Load the ia_source_withLandmarks. initial.coord in the REF configuration
> and the ia_target_withLandmarks_smoothed.coord as the AUX configuration,
> then toggle between REF and AUX views to see the starting configuration
> and the desired target configuration. Select Paint Control: Lobes to
> display the landmark nodes in color. Once the deformed sphere has been
> registered, the original spherical coordinate file is projected to the
> target sphere. This resultant deformation-map file allows data from the
> source to be mapped to the corresponding location in the target sphere.
> Hope this helps,
> > Hi,
> > I have a question about the spherical registration mehtod you used. I
> searched a lot, and found that the landmark-based spherical registration
> method you use is developed by Bakircioglu M, Joshi S, Miller MI,
> "Landmark matching on the
> > sphere via large deformation diffeomorphisms". Proc SPIE Med
> > Image Processing. 1999;3661:710?C5.
> > I found this reference in the paper " An Integrated Software Suite for
> Surface-based Analyses of Cerebral Cortex" talking about Caret. So my
> question is, is the method developed by Bakircioglu still the one
> implemented in Caret for the spherical registration? As this method was
> developed in 1999, I'm not sure whether there are any new updates or
> changes for it. It's quite hard to find any reference about the method
> in your newest papers.
> > Hope for your hints.
> > Thank you very much.
> > Jidan
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the caret-users