Antwort: Re: [Insight-users] Re: Registration Slice to Volume

[matthias.honal at uniklinik-freiburg.de]

Hi,

I'm currently looking into slice-to-volume registration as well, but I
encountered a problem with the suggestion to use the slice as fixed image
in my particular case.

I have one reference volume and multiple slices which are _non-rigidly_
deformed with respect to the reference volume. My goal is now to register
all these slices (one after the other) to the reference volume and not vice
vera, i.e. in my case the volume _has_ to be the fixed image, and each
slice a moving image.

Now, for the ITK registration framework that does not make sense as Rupert
and Luis wrote, since one would run into problems with the derivative of
the image metric (a 3D gradient would have to be computed from a 2D image).
In case of a rigid transformation one could easily register the volume onto
the slice and then invert the transformation, however deformable
transformations (such as  BSpline transformations) are not invertable in
general as far as I know.

So my question is what could be done in this case.
- Look for an invertable, deformable transformation? (I know that there are
some, which may be pretty slow however.)
- Implement an alternative to the way registration is currently done in
ITK. That would mean to map points from the moving image onto the fixed
image and then find some method to "distribute" a pixel value from the
moving image to its neighbors on the fixed image pixel, when a moving image
point is mapped "between" the points on the fixed image grid.

Are there any comments, suggestions or even already known ITK
implementations for these problems?

Thanks a lot,

Matthias




                                                                           
             Luis Ibanez                                                   
             <luis.ibanez at kitw                                             
             are.com>                                                   An 
             Gesendet von:              rbrooks at cyberus.ca                 
             insight-users-bou                                       Kopie 
             nces+matthias.hon          insight-users at itk.org              
             al=uniklinik-frei                                       Thema 
             burg.de at itk.org            Re: [Insight-users] Re:            
                                        Registration Slice to Volume       
                                                                           
             05.07.2007 19:09                                              
                                                                           
                                                                           
                                                                           
                                                                           





Hi Albert,

I second Rupert's opinion here.

It would be more natural to use the Slice as the fixed image.
Even though that may sound counter-intuitive.

You may want to read the introduction section of the
"Image Registration" chapter in the ITK Software Guide:

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


where this issues are discussed.


    Regards,


       Luis


-----------------------
Rupert Brooks wrote:
> Hi Albert,
>
> To help a bit more i think we would need to know more specifically
> what problems you have with the metric.
>
> I'd say that the slice has to be the fixed image, otherwise there will
> be problems computing the derivative in the direction normal to the
> slice.  I'd also recommend using a different interpolator.  The
> Nearest neighbor can pose problems for registration as it leads to a
> very rough cost function.  A linear interpolator is also far from
> ideal, but usually manages to work anyway.
>
> Rupert B.
>
>
>> Jul 2, 2007 (yesterday)
>> Registration Slice to Volume
>> from ITK by Albert Gubern
>> Hi,
>>
>> I'm trying to do Registration slice to volume. Firstly and to make it
>
> as easy as >possible, I extract a slice (fixedImage) of the volume 3d
> (movingImage). I want the >registration say me the z of the slice into
> the volum 3D .
>
>>
>> Slice is casted to volume of only one slice to have two image with
>
> the same >dimension.
>
>>
>> typedef itk::Image<unsigned char,2> ImageType;
>> typedef itk::Image<float,3> FixedImageType;
>> typedef itk::CastImageFilter< ImageType, FixedImageType >
>> CastFilterType2D;
>> caster2D->SetInput( filter->GetOutput() ); //filter->GetOutput() is
>
> the slice 2D
>
>>
>> Moreover, I've tried it with a TranslationTransform,
>> MeanSquaresImageToImageMetric and
>> NearestNeighborInterpolateImageFunction but I have problems with
>
> metric.
>
>>
>> I think that this is the way to make it, but I am not sure if I have
>
> to make a >different cast or if I have to choose another metric.
>
>>
>> I'll thank if someone can help me, say me where can be the mistake or
>
> which >metric and interpolator I have to use.
>
>>
>> Cheers
>>
>> --
>> Albert
>
>
_______________________________________________
Insight-users mailing list
Insight-users at itk.org
http://www.itk.org/mailman/listinfo/insight-users