[Insight-users] Re: Image or Mesh to Tubes

[Julien Jomier]

Kevin,

In our case we have a predefined idea of the branching pattern since the 
vasculature is segmented from the pre-operative data and then aligned.

In your experiment, do you think you can form a generic model of the 
dendrites and fit that to your data in order to compare them?
The other option is to perform a segmentation task on each image and do 
a model-to-model comparison. I'm not sure about the quality of your 
images though. There are some tubular segmentation algorithms in ITK 
(look at the itkCurvesLevelSetImageFilter) but none of them produces a 
TubeSpatialObject as output and this will require some post-processing.

Registration has some advantages v.s. segmentation in the sense that it 
establishes correspondences inherently (but it can also establish false 
correspondences). Also registration is sometimes less sensitive to noise 
and missing data.

Julien

Kevin H. Hobbs wrote:
> On Wed, 2006-08-02 at 11:13 -0400, Julien Jomier wrote:
>> Do you plan to use a rigid or deformable transformations?
> 
> I don't think we are far enough along that I can rule out either.
> 
>> Are you interested in model-to-image registration or do you investigate 
>> model-to-model registration as well?
> 
> I think I have to say both. The idea is to fit tubes onto the images,
> and then to compare and contrast the produced models. 
> 
> We have many 3D confocal images of electrically identified lobster STG
> neurons that were injected with a fluorescent die. So we know the cell
> that appears in each image by name. So far we don't have models for what
> the cells look like, other than "The dendrites look like branching
> trees." In fact the present idea is that the branching pattern is not
> genetically determined and so it is totally random. We suspect that this
> is not really true and we should be able to find something that
> distinguishes them.
> 
> 
>> In our approach we first do a global rigid alignment then a piece-wise 
>> rigid alignment of each branch starting from the trunk, then we perform 
>> a fully deformable registration.
>>
> 
> Do you have a predefined idea of what the branching pattern looks like?
> 
>> More information can be found about in these papers:
>>
>> Aylward S, Jomier J, Weeks S, Bullitt E, "Registration of Vascular 
>> Images," International Journal of Computer Vision, March 2003, pages 15
>>
> 
> I can get this one and I'll give it a read.
> 
>> Jomier J, Aylward S, "Rigid and Deformable Vasculature-to-Image 
>> Registration: A Hierarchical Approach", MICCAI 2004
> 
> I can't get this one.