[Insight-users] RE: [Insight-developers] icosahedron

[Alexandre GOUAILLARD]

hi everyone.

leila,

1. there is no problem to import vtkPolydata to itkMesh, especially if 
the polydata is as regular as the output of vtkPlatonicSource. You can 
use the filters provided in itkApplications, more specifically 
"vtkPolydataToItkMesh" project.
2. You can indeed use the itkQE code which also provide several vtk=>itk 
  importers (polydata and UGrid). then you would have a better structure 
to work with. To begin with you surface could only be 2-manifold, which 
is quite convenient for brain segmentation, isn't it?
BTW: we would need reviews so that would be a good opportunity to get 
two birds with one stone ;)
3. we plan to import many filters from vtk to itk, but we haven't 
discussed about sources right now. Sources are usually used only once 
(and mesh can be stocked) whereas filters have a more important role in 
pipelines, thus the priority given to filters.

As for the question of initialisation of snakes, in all our 
experimentations, when using trinagle meshes, a regular source is always 
  preferable. The vtkMeshSource as two singular vertices for which the 
valence is way higher than for the other points, and most deformations 
algorithm (and below this, curvature approximation algos) are known to 
be very unstable when valence is not 6.
We had in practice the best results with an hybrid algorithm, very simple:
= start from a regular icosahedron (possibly subdivided to match the 
sampling you require).
= start loop
    - apply your deformation algorithm
    - if an edge is getting longer than a threshold subdivide both faces
      irregularly (i. e. 1 to 2, or 1 to 3, not 1 to 4).
    - if the quality factor of the triangle is too low, flip an edge.

the irregular subdivision allow you to control the sampling: if you have 
one part of the mesh getting into a long sock-like part, you need to 
adapt locally the sampling/connectivity of the mesh to achieve good results.
the flipping is a cheap way to keep your mesh quality high. Algorithms 
are often sensitive to that.

Hope this helps.

Luis Ibanez wrote:
> 
> Hi Leila,
> 
> The itkRegularSphereMeshSource that Thomas contributed to ITK
> starts with six points equidistant in a unit sphere.
> 
> I wonder if the first subdivision of that figure will give you
> an icosahedron... (maybe not...)
> 
> 
> In any case,
> Do you really need an icosahedron ?
> or just a simple figure to initialize a deformable surface ?
> 
> 
> 
> Also,
> you may want to look at the recent paper in the Insight Journal:
> 
> http://www.insight-journal.org/InsightJournalManager/view_reviews.php?back=publications.php%3Fjournalid%3D6%26order%3Drating%26revision_display%3Dcombined&pubid=122 
> 
> 
> itkQuadEdgeMesh:
> A Discrete Orientable 2-Manifold Data Structure for Image Processing
> 
>        http://hdl.handle.net/1926/306
> 
> 
> 
> 
>    Regards,
> 
> 
>       Luis
> 
> 
> 
> ----------------------
> Leila Baghdadi wrote:
> 
>> Hello Jim,
>>
>> Thanks for the reply.
>>
>> I am wondering if it is a good idea to actually create a cell topology
>> class for icosahedron in itk, I guess itk only has TetrahedronCell.
>>
>> Sylvain also mentioned that vtk has Platonic solids with the 5 solids
>> including icosahedron however I am not sure if we have any code which
>> can convert this to an itk mesh.
>>
>> I am writing some code for brain segmentation and have to do this as
>> part of my work, so I am just trying to determine what is the best
>> strategy at this point.
>>
>> any thoughts
>>
>> Thanks again for your time
>>
>> Leila
>>
>> On Mon, 2006-18-09 at 14:08 -0400, Miller, James V (GE, Research) wrote:
>>
>>> Leila,
>>>
>>> I thought the SphereMeshSource provided an icosahedron but looking at
>>> the API, it looks more like spherical coordinates parameterization.
>>>
>>> I think http://insight-journal.org/dspace/handle/1926/215 has a
>>> SphericalHarmonicMeshSource which generates an icosahedron.
>>>
>>> Jim