[Insight-users] Control of deformation field generation via landmarks?

[Alexandre GOUAILLARD]

Ruben,

If you're going to implement such a filter, you might want to take a look at
the two following families of filters:

Distance Maps - 
http://www.itk.org/Doxygen/html/classitk_1_1ApproximateSignedDistanceMapImag
eFilter.html

GradientVectorFlow -
http://www.itk.org/Doxygen/html/classitk_1_1GradientVectorFlowImageFilter.ht
ml

I guess what you want to do (extend a sparse vector field to the entire
image space) is closer to Gradient Vector Flow than of Distance Maps, but
practically  (propagation of a sparse value over the image space) you should
be able to get insight from both filter's families (and there are more
implementation of distance map than vector flows).

I guess the pseudo code would look like that:
* Create your initial (sparse, only within range of the landmarks) vector
field,
* extend/extrapolate/possibly smooth/ it
* displace
* compute whatever metric you are using
* iterate

Regards,

Alex "HanFei" Gouaillard.

On 7/11/07 12:20 PM, "Ruben Schilling" <r.b.schilling at googlemail.com> wrote:

> Hi Luis,
> 
> I tested actually all 5 Kernel transforms in ITK for this and I
> discovered my problem actually is, that there are no landmarks on the
> border of the image. Thus the splines fold however they like to fold
> outside of the small landmarks regions. The interpolation near my
> landmark positions looks very good though.
> I guess what I need is a more or less isotropic copy process (maybe
> with slight averaging) of the few displacement vectors I have to
> cover all of the image foreground. Is there a filter to do an
> isotropic masked extrapolation of a sparse deformation field? If not
> I probably need to write one myself.
> 
> Regards,
> Ruben
> 
> 
> 
> Am 11.07.2007 um 18:44 schrieb Luis Ibanez:
> 
>> Hi Ruben,
>> 
>> The KernelTransform used inside this filter *already* blends
>> the deformation from one landmark to the next.
>> 
>> You may want to play with replacing the reciprocal elastic body
>> kernel with the elastic body one.
>> 
>> If you want to see profiles of the different kernels you can
>> use the application:
>> 
>>         InsightApplications/ThinPlateSplines
>> 
>> 
>> You should also play with the value of Stiffness. This parameter
>> allows to modify the behavior of the KernelTransform from
>> interpolation to approximation. The default behavior is to
>> interpolate.
>> 
>> 
>> When applying deformations it is wise to add landmarks on the
>> outside of the image, and to have their homologous landmarks
>> to be superimposed, so that they generates a constrain of
>> zero deformation around the image.
>> 
>> 
>> 
>> Regards,
>> 
>> 
>>     Luis
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users