About the data used to build the shape model

[Robin Huang]

Hi,

In the ITK examples, the shape model was built from a set of vtk poly data. My question is how do you create these poly data? I have a a whole bunch of segmented data which I want to use to construct a statistical shape model. What do I need to do to turn them into the necessary data that statismo requires to create the shape model?

Thanks

Robin

[Thomas Albrecht]

Hi Robin,

you will need to "register" the data. Are you familiar with the concept of non-rigid image (or shape) registration?

If not, I can elaborate a bit. For now, statismo does not include its own registration algorithm. You can for instance use elastix ( http://elastix.isi.uu.nl ) to register your data.

I hope that helps

Tom

--
You received this message because you are subscribed to the Google Groups "statismo" group.
To unsubscribe from this group and stop receiving emails from it, send an email to statismo-user...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.

[Robin Huang]

Hi Tom,

I think you might have misunderstood my question. I am asking about the file formats used during the construction of the shape model. Not about using the built shape model to perform registration on unsegmented image data. Either that or I am seriously not getting the entire process of model based segmentations. 

Just to clarify if I am on the right track, with model based segmentation approaches, you use a whole bunch of pre-segmented (training) data to construct either an atlas or statistical shape which you then use to segment other images, and registration is used in the second part of the process, right? I am really new to whole medical image analysis and processing stuff, so I am not confident at all if I understood everything. 

Anyway, what I was asking was; so I have a whole bunch of pre-segmented training images, how do I use statismo (or any other necessary tools) to create the shape model? 

Robin

[Thomas Albrecht]

Hi Robin,

you are correct in your statement that the segmented images are used as training data for the model, which is then used for registration of novel data sets.

However, to build the model in statismo, the training data needs to be in correspondence. This means, they need to be registered. This goes for practicallly any other statistical model in the literature, not just for statismo. There are a couple of papers building shape models of distance maps. There, no registration is necessary. But I would not recommend these methods readily.

As a really good paper for reading up on model-based segmentation, including the initial registration of the training data, I can recommend the review paper by Heimann and Meinzer: http://scholar.google.com/scholar?cluster=9638374698290510469&hl=de&as_sdt=0,5 .
Shape correspondence is handled in Section 4.

As for the actual registration, I would recommend first a landmark-based rigid alignment of your data. Then I would compute a distance image for each of the aligned shapes, this is a convenient and well-behaved representation of the shapes. Then I would register these distance images with elastix. It takes a bit of effort to do this and to find the right parameters for elastix.

I hope this helps you

Tom

[Ibrahim A]

Hi,

I feel I am missing something here, the original question is still not answered. how to convert the image data e.g. vtk or png to vtk polydata? or maybe there is another way to use them directly without converting? 

[Marcel Luethi]

Hi,

The first step to convert an image into a vtk polydata is to segment
the structure of interest from the image. The result of this
segmentation is a binary image where the voxels that belong to the
structure of interest have the label 1 and all the other voxels a
value of 0. From this binary image a vtkPolyData can be generated
using for example the vtkMarchingCubes algorithm
http://www.vtk.org/doc/nightly/html/classvtkMarchingCubes.html


Best regards,

Marcel

> For more options, visit https://groups.google.com/d/optout.

[Ibrahim A]

Hi Marcel,

sorry for the late comment, I was in a vacation. Thanks for the detailed answer. The images are already segmented. I will try vtkMarchingCubes. 

Best regards,

Marcel

> email to statismo-users+unsubscribe@googlegroups.com.