[Insight-users] segmentation reconstruction of the tongue

[Luis Ibanez]

Hi Michael,

We had a long conversation this morning at SPIE
wandering about how you are acquiring this tongue
data and what are you planning to do with it.

 From this last email it seems that you are not
acquiring volumes but a single axial slice and
a single saggital slice. And these pair of
slices are acquired for different times during
the movement of the tongue.

Is that right ?

Given the natural symmetry of the tongue I would
suggest you to work first with the saggital series
of 2D slices because this series is always capturing
the same plane of tissue, while the axial slice will
be intersecting a different plane of the tongue as
it moves.

And easy first exercise for you is to use the Demons
registration methods and try to do deformable registration
between two saggital slices in consecutive times.

You can do this with the example in

    Insight/Examples/Registration/
             DeformableRegistration2.cxx

You can give directly as command line arguments the
filenames of the two consecutive saggital slices.
(using the DICOM filenames).


Demons will compute a deformation field, and then the
ResampleImageFilter will resample the second slice in
order to fit it on top the first one.

For you,
the interesting aspect is to look at the deformation field.
You can do this with the open source application ParaView


         http://www.paraview.org


If you take your sequence of saggital slices, let's
call them

             s(1), s(2), s(3),....s(N)

and using Demons by pairs you compute deformation fields

            d(i) = Demons( s(i) , s(i+1) )

then you can take all this deformation fields and run
an animation in ParaView. That will show you the evolution
of the deformation field as the tongue moves.


Note that Demons will work better if the tonge movement
from one slice to the next is not creating displacements
of more that a couple of pixels.

If you find that the deformations are larger than that
you may want to use landmarks for initializing the deformation
field and pass this initialization to the Demons filter.

Please look at the example of manual landmark initialization:


         Insight/Examples/Registration/
                             LandmarkWarping2.cxx


---

You can also try the FEM-based deformable registration method,
by simply running the example:


      Insight/Examples/Registration/
                DeformableRegistration1.cxx

Please read the description of the parameters file in the
SoftwareGuide


       http://www.itk.org/ItkSoftwareGuide.pdf



Regards,



    Luis



-------------
a a wrote:

> Hi,
>  
> my MRI data of the tongue (with tagging) comes in two orthogonal planes, 
> one for short axis and one for the long axis.
>  
> In order to get a 3D deformation of the tongue, information from both 
> short and long axes tagged lines have to be utilized. I'm wondering if i 
> should just segment and reconstruct the short axis and long axis 
> separately or together? Are there any advices as in what i should do or 
> use for this problem?
>