[Insight-users] Similarity3Dtransform: Scale value interpretation

[Luis Ibanez]

Hi Emma,


Question 1:

When the Similarity3DTransform gets a negative value in the Scale
parameter, that performs a reflection of the space on the origin
of coordinates.... not a good thing at all, particularly in medical
images.

The scale shouldn't approach a zero value either...

This however, is possible when you use an optimizer that assumes
that the parameters of the transform are in a Vector space, as
most of the GradientDescent optimizers do.


In practice you could have prevented the scale from deforming that
far by setting proper values in the parameter scaling array that
you pass to the optimizer. The purpose of this array is to let
the optimizer know that we don't expect big changes to occur in
the Scale parameter of Transform.



Question 2:

You should only need a Similarity transform is you are registering
images from different patients, that will naturally have slightly
different organ sizes. Note however that the size variability of
human specimens is not as large as to need scales smaller than 0.5
or larger than 2.0.

If you are registering two different patients, then, in addition
to having the origin & spacing taken into account, you may also
need this type of scaling in space.


Question 3:

Resampling will take the image spacing into account. You just
need to make sure that the spacing is set correctly in the images.
Note that the resampling also use the transform anyways.


Question 4:

No, you don't need to manually compensate for the resolution.
ITK already takes the pixel spacing into account. All the
process of image registration is done in physical coordinates.


Question 5:

Yes, the Patient position is the source of the data for the
image origin values.




    Regards,


       Luis



-----------------
Emma Ryan wrote:
> Hi,
> 
>   I am using the itk Similarity3DTranform to obtain a registration 
> between two volumes.
> Dataset1 : 512 x 512 x 30 (Resolution = 0.684 x 0.684 x 1.0) :  Image 
> Patient Position : -175, -175, 125 (from dicom header)
> Dataset2: 512 x 512 x 30 (Resolution = 1.0 x 1.0 x 1.0 ) : Image Patient 
> Position: -255, -255, 101 (from dicom header)
> 
> Both are of the same modality (CT).
> 
> I have a few questions.
> 
> 1. How does one interpret the scale value of  (-0.00008). What does 
> negative mean ?  Division ?
> 
> 2. Prior to registration I mention the size, origin and spacing for both 
> the fixed and moving volumes. At resampling, the moving volume are 
> resampled to match the spacing and origin of the fixed volume. So should 
> I care to include the scale transform in my registration/optimization 
> parameters ?
> 
> 3. Does resampling take care of the resolution difference ?
> 
> 4. Is it necessary that I set the initial scaling to 1.428 to compensate 
> for the resolution ?
> 
> 5. Dicom images have Image Patient Postion tag. Is this the information 
> that goes into the  "Origin" variable ?  If not, where do I get the 
> image origin information ?
> 
> I have read a lot of literature on resampling and all the pdfs and ppts 
> on ITK, but I fail to understand these.
> 
> Please help !
> 
> Emma
> 
> ------------------------------------------------------------------------
> Get the Yahoo! toolbar and be alerted to new email 
> <http://us.rd.yahoo.com/evt=48225/*http://new.toolbar.yahoo.com/toolbar/features/mail/index.php>wherever 
> you're surfing.
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users