[Insight-users] Masks & registration

[Luis Ibanez]

Hi Laurent,

The Transform Initializers do not take into account the Masks.

They compute the Moments/Geometry on the entire images.

Note that a better way to initialize your registration is to
use the LandmarkBasedTransformInitializer.

http://www.itk.org/Insight/Doxygen/html/classitk_1_1LandmarkBasedTransformInitializer.html

It requires the user to provide at couple of homologous points
in both images, but it will increase a lot the quality of your
initialization.



   Regards,


      Luis



======================
Laurent PAUL wrote:
>  Hi users,
> 
> I've got a problem with masks & registration.
> I use MeanSquaresImageToImageMetric and VersorRigid3DTransform for the
> registration process.
> Since I have regions of interest, I tried to pass at least 1 mask to the
> metric. It seems to be ok as I test it with IsInside() method with two
> points (one in and one out).
> My first question is if I use SetDefaultOutsideValue(0) for the mask, any
> value different from 0 will be considered as Inside, right?
> 
> 
> Then, I read a message which deals with nearly the same problem:
> http://public.kitware.com/pipermail/insight-users/2005-September/014948.html
> 
> This is an (interesting) extract:
> 
>>"Sure, as you pointed out, if you start off with a horrible initial
>>misalignment, in the worst case, the regions don't overlap and the metric
>>never gets computed.  Usually, in a registration application, if you've
>>gone through the trouble of specifying relevant regions in both the
> 
> fixed >and moving image regions, you already have some idea of the
> initial >transform and there is no excuse for not specifying an initial
> transform >that guarantees significant overlap between the two masks.
> 
>>(This will of course only happen if you've specified *both* a fixed and
>>moving image mask, and not if just specify one of them. )"
>>
> 
> 
> I am in this case, where I know regions of interest.
> I understand the intialization problem but how initialize the
> registration?
> So far, I used the CenteredVersorTransformInitializer:
>    initializer->SetTransform (transform);
>    initializer->SetFixedImage (FixedImage);
>    initializer->SetMovingImage (MovingImage);
>    initializer->MomentsOn();
> 
> What I understood is that initialisation is made upon Mass center. Is mass
> center calculation takes account of the mask, I mean only region contained
> in the mask?
> I tried too with initializer->GeometryOn();
> but I've got an error at runtime. Is it safe to use "GeometryOn"?
> 
> 
> More, I know initialisation is quite ok. There is just a translation error.
> I tried (without using Initializer):
>   TransformType::OutputVectorType InitialTranslation;
>   InitialTranslation[0] = 80;
>   InitialTranslation[1] = 0;
>   InitialTranslation[2] = 0;
>   transform->SetTranslation( InitialTranslation );
> Registration's result is bad and stops. I saw that transform center is
> [0,0,0].
> So, I first use intializer (to compute transform center) and
> OverrideTranslation with mine.
> Is there a better way to initialise the registration?
> 
> 
> If I specify a MovingMask, even with good initialization, registration
> stops with this error message: All the points mapped to outside of the
> moving volume.
> So, my last question is: Is mask initialized with the initialtransform?
> 
> Sorry for the long message but I had to be clear.
> 
> Laurent.
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 
>