[ITK-users] Own Transformation Class

[Matt McCormick]

Hey Dominik,

> A new transformation can often be avoided by calling SetWeights() on the
> > optimizer as explained here [1]. This can be used to hold parameters
> > constant for a given transform during the optimization process.
> >
> > However, it sounds like goal is to limit registration to information from
> > a particular region in the image (the urethra). To achieve this goal, you
> > may consider setting the Fixed and Moving image masks on the metric as
> > described here [2].
>
> I'm already using registration masks on the prostate itself to exclude the
> surrounding tissue. However, the registration is always a bit off because
> in
> the result the urethras are not in the same place. In the histology the
> urethra is more or less a point and in the MRI the urethra is like a vector
> (well actually it is more like a curve). So I'd like the transformation to
> edit it in a way that the point follows that curve. I hope my explanation
> is
> not too confusing :)
>

In this case, an approach that uses the corresponding fiducials with the
LandmarkBasedTransformInitializer:


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

or a registration that combines a point set metric in a multi-metric may be
helpful.


http://www.itk.org/Doxygen/html/classitk_1_1PointSetToPointSetMetricv4.html


http://www.itk.org/Doxygen/html/classitk_1_1ObjectToObjectMultiMetricv4.html


> If every histology slice is registered independently, when they are merged
> > together there will be gaps and overlap in the histology slice locations.
> > Therefore, if a histology volume is desired, the histology slices could
> be
> > registered to each other first to generate a single histology slice
> > volume, then that volume is registered to the MRI volume.
>
> The histology slices all have a certain distance to each other (and are not
> even parallel I think), so I think gaps in the merged volume are desirable.
> Therefore I can't just stack them.
>

As Andras mentioned in another thread, this is a hard problem. A basic
first step could add images together generated with ResampleImageFilter.
The volume reconstruction executable available in PlusLib is also nice.

HTH,
Matt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/insight-users/attachments/20151204/f43db642/attachment.html>