[Insight-users] SetFixedImageRegion : Demons & constrast image registration

Luis Ibanez luis.ibanez at kitware.com
Fri Oct 8 12:36:02 EDT 2004 

Hi Mark,

Thanks for letting us know that the Extract+Demons
combination worked for you.


1) The issue with usind Demons for registering
    pre versus post constrast images is that Demons
    attempt to shift tissue in order to match the
    intensity levels between the images.  There is
    a risk that the presence of the contrast agent
    may induce mis-matches between both input images.

    When I suggested to use a mask over the contrast
    agent region I was assuming some sort of angiography
    images where the contrast agent is usually constrained
    to a relatively small region of the image.

    However, if you are in a case where the constrast
    agent can reach almost any corner, then you probably
    have to go for a Mutual Information metric and one
    of the two following methods


    A) FEM deformable registration

    B) BSplines deformable registration



2) Smoothing may not help you in this case in the
    Demons methods. Adding more smoothing is equivalent
    to increading the rigidity of the deformation. It
    doesn't really affect the potential intensity mismatches.
    The more you smooth the deformation field, the more
    uniform the deformation field becomes. This is a trade-off
    between regularization (noise reduction) and resolution
    of the deformation field.



3) The method that you found in the Tutorial is explained
    in detail in the examples:

      Insight/Examples/DeformableRegistration4.cxx
      Insight/Examples/DeformableRegistration6.cxx

    This uses a BSplineTransform and the LBGSB optimizer.
    In this configuration you can select any of the ITK
    image metrics.



Please let us know if you have any further questions,



    Thanks



      Luis




-------------------------
Mark Wyszomierski wrote:
> Hi Luis,
> 
> Thanks for the tip, I used the ExtractImageFilter to hand the Demons
> Deformable registration filter a region of interest, and it worked
> great.
> 
> I am trying to register a few different organs, they are breast,
> prostate, and heart, all using contrast agents. You suggested using a
> masked registration in which the area that intakes the contrast agent
> is not considered, but for the breast especially it seems there are no
> other landmarks to register off of that are not affected by the
> contrast agent, if I understand correctly?
> 
> If this is the case, should I just attempt to increase smoothing in
> order to more effectively use the demons registration? What are the
> consequences of increasing the smoothing?
> 
> Lastly, I was reading the ITK tutorial presentation on registration,
> slide 64 shows an MRI breast study with contrast enhancement
> registered - I'm wondering what method was used to accomplish this?
> The caption below the images reads:
> 
> Registration: 2 levels, MI, LBFGS, B-spline deformation
> 
> Thank you so much Luis,
> Mark
> 
> On Thu, 30 Sep 2004 16:59:22 -0400, Luis Ibanez <luis.ibanez at kitware.com> wrote:
> 
>>Hi Mark,
>>
>>There is not a direct way to specify a region of interest to the
>>Demons Deformable registration filter. However, you can easily
>>obtain the equivalent effect by extracting a region of interest
>>from the input images using one of the following filters
>>
>>   --- ExtractImageFilter
>>http://www.itk.org/Insight/Doxygen/html/classitk_1_1ExtractImageFilter.html
>>
>>   --- RegionOfInterestImageFilter
>>http://www.itk.org/Insight/Doxygen/html/classitk_1_1RegionOfInterestImageFilter.html
>>
>>"Consensus" is unfortunately a very uncommon word in medical image
>>processing. Our field is still in the pre-adolescent stage where
>>everybody wants to demonstrate that their method is better than the
>>others. We are still caring more for our reputation than for the
>>well-being of patients, so unfortunately the answer to your question
>>will be different depending on who you ask.
>>
>>What I will suggest you is to use a "Masked" registration. That should
>>matter more than the selection of one transform over another. You may
>>want to setup a registration in which the region that will intake the
>>contrast agent will *not* participate in the metric computation of
>>the registration process.
>>
>>The Demons method may work fine if you apply enough smoothing, but
>>there is a tradeof between how much you want to smooth...
>>
>>What anatomical structure are you working on ?
>>
>>If it is the brain, you probably want to stay with a rigid registration
>>and verify how many image features are remaining in the difference
>>image after registration.
>>
>>Please let us know if you have further questions.
>>
>>   Thanks
>>
>>     Luis
>>
>>----------------------------
>>
>>
>>Mark Wyszomierski wrote:
>>
>>
>>>Hi Luis,
>>>
>>>That was it and it worked great, thanks! Is there a way to hand a
>>>region of interest to the demons deformable registration method? I am
>>>basically using the example in the itk example folder,
>>>(DeformableRegistration2).
>>>
>>>Also, is there some consensus on which registration method is best for
>>>an MR dataset using contrast agents? The rigid registration seems to
>>>work fairly well, the demons deformable has worked wonders though in
>>>other registration problems I have handed it. But by the nature of the
>>>demons registration (intensity matching) this is probably not feasible
>>>for MR contrast enhanced datasets). Any info would be appreciated,
>>>thank you very much Luis,
>>>
>>>Regards,
>>>Mark
>>>
>>>
>>>On Fri, 24 Sep 2004 13:55:36 -0400, Luis Ibanez <luis.ibanez at kitware.com> wrote:
>>>
>>>
>>>>Hi Mark,
>>>>
>>>>You can define your region of interest using
>>>>lines like the following (assuming 3D...):
>>>>
>>>>ImageType::IndexType start;
>>>>start[0] = 144;
>>>>start[1] = 123;
>>>>start[2] = 223;
>>>>
>>>>ImaggeType::SizeType size;
>>>>size[0] = 400;
>>>>size[1] = 400;
>>>>size[2] = 400;
>>>>
>>>>ImageType::RegionType region;
>>>>
>>>>region.SetIndex(start);
>>>>region.SetSize(size);
>>>>
>>>>registration->SetFixedImageRegion( region );
>>>>
>>>>It will be up to you to define the actual values
>>>>to put in "start" and "size" in order to define
>>>>the region of interest. They should probably be
>>>>taken from a GUI interace by letting the user do
>>>>mouse clicks....
>>>>
>>>>Just make sure that the region is entirely contained
>>>>inside the BufferedRegion, otherwise the Metric will
>>>>throw an exception.
>>>>
>>>> Regards,
>>>>
>>>>    Luis
>>>>
>>>>-----------------------
>>>>Mark Wyszomierski wrote:
>>>>
>>>>
>>>>
>>>>>Hello All,
>>>>>
>>>>>I have been using motion registration and it is great. A quick question -
>>>>>
>>>>>Currently I am considering the whole image for registration, when
>>>>>using only a region of interest would be preferable. Currently I am
>>>>>using this method to set the considered region:
>>>>>
>>>>>registration->SetFixedImageRegion(
>>>>>fixedItkBuffer->GetOutput()->GetBufferedRegion() );
>>>>>
>>>>>fixedItkBuffer is a buffer containing the pixel data for a reference
>>>>>image. registration is my registration type pointer.
>>>>>
>>>>>How then can I specify a simple rectanglar region of interest to be
>>>>>considered instead of the entire buffer?
>>>>>
>>>>>Thanks!
>>>>>_______________________________________________
>>>>>Insight-users mailing list
>>>>>Insight-users at itk.org
>>>>>http://www.itk.org/mailman/listinfo/insight-users
>>>>>
>>>>
>>>>
>>>_______________________________________________
>>
>>
>>>Insight-users mailing list
>>>Insight-users at itk.org
>>>http://www.itk.org/mailman/listinfo/insight-users
>>>
>>>
>>
>>
> 
>

More information about the Insight-users mailing list