[Insight-users] Re: [Paraview] 3D Deformable Registration Deformation Field

[Luis Ibanez]

Hi Kevin,

Thanks for letting us know that you found the source
of the error in the voxels spacing of your datasets.

Regarding the validation, unfortunately there are no
standards on how to validate deformable registration.

The common approach is to artificially generate a
deformation field by using ThinPlateSplines, use this
artificial deformation field in order to create a
deformed version of your image. Then use your registration
method between the original and the deformed image, and
finally compare the artificial deformation field against
the one that you produced with the registration method.

Another option is to compare the resampled deformed image
against the original image (via a squared image difference).

 From the pragmatic point of view, I would not expect
to be able to recover a deformation field because in
most cases the problem is not invertible. In any case,
for a medical application it is worth to design some
verification test.



    Regards,


       Luis



------------------
Kevin Ming wrote:
> Hi Luis,
> 
> The fixed and moving images did indeed have voxel dimensions of 
> 0.02x0.02x27.00 mm.  This, however, is NOT correct.  So I went in and 
> changed the header files for my images to the proper dimensions, 
> performed registration, and obtained the field.  It looks much better 
> now :)
> 
> This is probably a dumb question, but are there ways using ITK to 
> validate my deformation field result?
> 
> 
> Thank You,
> Kevin
> 
> On 7/11/06, * Luis Ibanez* < luis.ibanez at kitware.com 
> <mailto:luis.ibanez at kitware.com>> wrote:
> 
> 
> 
>     Hi Kevin,
> 
> 
>     Thanks a lot for sending the MetaImage header.
> 
>     The line:
> 
>          ElementSpacing = 0.02 0.02 27
> 
>     explains what we are seeing in ParaView.
> 
> 
>     The voxel size is about 1000 times larger
>     in the Z direction     :-/
> 
> 
>     ParaView is showing the reality of the data.
> 
> 
> 
>     What you may want to verify now, is whether that
>     was the original pixel size of the images that
>     you were registering. In particular you should
>     look at the Fixed image.
> 
>     Note that if that's the real spacing on the fixed
>     image... well... there is not hope for registering
>     such dataset, because it is a more of a rod than
>     it is a volume...   :-/
> 
> 
>     Please let us know what you find.
> 
> 
>         Thanks
> 
> 
> 
>            Luis
> 
> 
> 
>     -------------------
>     Kevin Ming wrote:
>      > Hi again Luis,
>      >
>      > I've uploaded both "set5o_s_10_field.mhd" and
>     "set5o_s_10_field.raw" to
>      > Kitware Uploader for your convenience.
>      >
>      > Also, for these, I used the Level Set Deformable Regsitration
>     with 500
>      > iterations.  I had to the extend it from the original 2D case to
>     3D, so
>      > I probably did something wrong when I changed the section of the
>     code
>      > for producing the deformation field, causing the problem I have
>     now.  I
>      > obtained a near-perfect registration, though.
>      >
>      > I hope this information is useful.
>      >
>      >
>      > Thank You,
>      > Kevin
>      >
>      >
>      > On 7/11/06, *Luis Ibanez* <luis.ibanez at kitware.com
>     <mailto:luis.ibanez at kitware.com>
>      > <mailto: luis.ibanez at kitware.com <mailto:luis.ibanez at kitware.com>
>      >> wrote:
>      >
>      >
>      >     Hi Kevin,
>      >
>      >     Thanks for posting the screenshots of the deformation field
>      >     visualizations.
>      >
>      >     Your verbal description was accurate.  Therefore, there is
>      >     definitely something wrong going on with the deformable
>      >     registration process.
>      >
>      >     Could you please post the .mhd files of your deformation
>      >     fields ?
>      >
>      >     like "set50_s_10_field.mhd"
>      >
>      >     Those are the text headers of the MetaImage format, they
>      >     should give us an idea of the number of pixels in the 3D
>      >     deformation field as well as the pixel size in 3D.
>      >
>      >
>      >        Thanks
>      >
>      >
>      >           Luis
>      >
>      >
>      >
>      >     --------------------
>      >     Kevin Ming wrote:
>      >      > Hi,
>      >      >
>      >      > I was able to get a deformation field output from my 3D
>     deformable
>      >      > registration, but it's not what I was expecting to see:  There
>      >     seem to
>      >      > be 64 sets of arrows, which is good because it corresponds
>     to the 64
>      >      > slices of MRI data I used.  But each of the sets of arrows,
>      >     supposedly
>      >      > from each of the slices, seem to originate from a single
>     point on a
>      >      > line.  I was expecting to see the field in a manner similar to
>      >     that of
>      >      > the 2D deformable registration examples, where arrows are
>     located
>      >     on the
>      >      > images in plane, showing what features have moved and
>     where they
>      >     have
>      >      > moved to.  Is it possible to get that in the 3D case?  Are
>     there
>      >     other
>      >      > ways of visualizing 3D deformation fields in ParaView?
>      >      >
>      >      > Note that my fixed, moving, and output images for the
>      >     registration are
>      >      > in the ANALYZE format (.hdr and .img files) and my
>     deformation
>      >     field is
>      >      > in the Meta image file format (.mhd and .raw files).
>      >      >
>      >      >
>      >      > Thank You,
>      >      > Kevin
>      >      >
>      >      >
>      >      >
>      >    
>     ------------------------------------------------------------------------
>      >
>      >      >
>      >      > _______________________________________________
>      >      > ParaView mailing list
>      >      > ParaView at paraview.org <mailto:ParaView at paraview.org>
>     <mailto: ParaView at paraview.org <mailto:ParaView at paraview.org>>
>      >      > http://www.paraview.org/mailman/listinfo/paraview
>      >
>      >
>      >
> 
> 
>