[Insight-users] Histoligical images registration

[Luis Ibanez]

Hi Delphine,


Thanks for preparing and posting the screen shots.


I guess that you are using the LFBGSBOptimizer:
http://www.itk.org/Insight/Doxygen/html/classitk_1_1LBFGSBOptimizer.html

If that's the case, you could use the bounds restriction
to prevent the BSpline nodes that are over that patch
(patch = the section that you want not to deform) from
having values different from zero.

Note that the ParametersArray of the transform, contains
the components of the deformation vectors associated to
every node of the BSpline grid. By limiting the values
of deformation of some of those nodes you can restrict
the deformation of some regions of the image.

The drawback is that if you want to have fine control
over the size and shape of the deformation, you may
have to increase the size of the BSpline grid, and that
has a linear impact on the computation time of the
registration.

You may want to start by (manually) drawing your
BSpline grid on top of the image, in order to get
a feeling of how precisely you could control the
deformation of the external boundaries.


---


An alternative approach is to use the FEM based
deformable registration and to define landmarks.

Another alternative approach is to modify the
LevelSetMotionRegistration
http://www.itk.org/Insight/Doxygen/html/classitk_1_1LevelSetMotionRegistrationFilter.html
where you could apply a mask to every iteration
of the deformation field computation.


You may also want to look at the following paper
in the Insight Journal:
http://insight-journal.org/dspace/handle/1926/549
with the following reviews:
http://www.insight-journal.org/InsightJournalManager/view_reviews.php?back=publications.php%3Fjournalid%3D8%26order%3Drating%26revision_display%3Dcombined&pubid=156



     Regards,


        Luis


-------------------------------
delphine.ribes at epfl.ch wrote:
> Hi luis,
> 
> You can have a look at two screenshot of the images at the following location :
> http://ltswww.epfl.ch/ltsftp/BeforeBspline.png
> http://ltswww.epfl.ch/ltsftp/AfterBspline.png
> 
> The red rectangle on the BeforeBspline.png images shows the external border
> which should not be deformed as a part of the brain is missing.
> The yellow rectangle on the AfterBslpine.png images shows the deformation. The
> internal organ as the hippocampe is too deformed.
> 
> So the region is located on the external border. To simplify the problem, I
> would like to find a way to not deformed the external border. I already have a
> mask of the images.
> 
> Thanks a lot for all the help
> Delphine
> 
> 
> Selon Luis Ibanez <luis.ibanez at kitware.com>:
> 
> 
>>
>>Hi Delphine,
>>
>>There are various ways of preventing a region
>>of the image from being deformed during the
>>registration process.
>>
>>The appropriate method will depend on the location,
>>size and shape of the region that you want to preserve.
>>
>>
>>
>>Questions:
>>
>>    What is the shape of this region ?
>>
>>    Is is a circular region in the middle of the image ?
>>
>>    Is it the external border of the image ?
>>
>>    Is it a rectangular region ?
>>
>>
>>If you could post an explanatory diagram in a public web site,
>>that will be great.
>>
>>
>>   Thanks
>>
>>
>>      Luis
>>
>>
>>
>>-------------------------------
>>delphine.ribes at epfl.ch wrote:
>>
>>>Hi Luis,
>>>
>>>Thank you for the answer. But I would like to ask you another question. I
>>
>>am
>>
>>>looking for a way to tell the registration process to not deformed a part
>>
>>of
>>
>>>the image.
>>>Right now I am using :
>>>Bspline transform
>>>LBFGS optmizer
>>>Linear interpolator
>>>and Normalized Correlation Metric
>>>
>>>Is there a way, using those parameters to tell not deformed this part ?
>>>
>>>Thanks a lot
>>>Delphine
>>>
>>>
>>>Selon Luis Ibanez <luis.ibanez at kitware.com>:
>>>
>>>
>>>
>>>>Hi Delphine,
>>>>
>>>>
>>>>   Your registration problem sounds quite interesting.
>>>>
>>>>   One possible way of preventing the holes in the histological sample
>>>>   from affecting the registration results is to take advantage of the
>>>>   Masks.
>>>>
>>>>   You can define Masks in the Fixed image and in the Moving image
>>>>   for selecting what pixels you want to participate in the
>>>>   registration. If you create masks where the holes are excluded
>>>>   from the mask, then the registration metric will be computed without
>>>>   considering such holes.
>>>>
>>>>   It seems that you have already look at the ITK Software Guide
>>>>
>>>>             http://www.itk.org/ItkSoftwareGuide.pdf
>>>>
>>>>   in particular section 8.11.3
>>>>   "Registration using masks constructed with Spatial Objects"
>>>>   pdf-page 464-466.
>>>>
>>>>   What you want to do is to create Masks where all the pixels
>>>>   corresponding to holes of the histological sample, are set to
>>>>   Zero, and the others are set to non zero.  Note that you could
>>>>   also take advantage of the mask to remove all the pixels outside
>>>>   of the mouse body to count on the registration process. It is
>>>>   usually advisable to leave a small border outside the body, as
>>>>   a way of making sure the that transition between body and air
>>>>   contributes to the registration process.
>>>>
>>>>
>>>>   The trick now, is to find an easy way of building the Mask.
>>>>   Hopefully the holes in your images are clear enough for using
>>>>   simple segmentation methods.
>>>>
>>>>   If now, we may need to take a look at your images in order to
>>>>   give you better advice.
>>>>
>>>>
>>>>
>>>>    Please let us know if you have more questions,
>>>>
>>>>
>>>>        Thanks
>>>>
>>>>
>>>>           Luis
>>>>
>>>>
>>>>
>>>>
>>>>------------------------------
>>>>delphine.ribes at epfl.ch wrote:
>>>>
>>>>
>>>>>Dear all,
>>>>>
>>>>>I am doing Image registration of histological images. My goal is to align
>>>>
>>>>brain
>>>>
>>>>
>>>>>mice in the same space coordinate to compare them. To do so I am using a
>>>>>reference volume.
>>>>>During the sectioning, some part of the brain is lost which create holes.
>>>>
>>>>To
>>>>
>>>>
>>>>>align volume, I am using 4 following registration : land mark, rigid,
>>>>
>>>>affine
>>>>
>>>>
>>>>>and a non rigid one which is Bspline. the last one is used to align
>>>>
>>>>internal
>>>>
>>>>
>>>>>organs. The alignment is good. But if the external border is cut,
>>
>>artifacts
>>
>>>>>appear.
>>>>>
>>>>>Do you know if there is a way to mask the hole to avoid those artifacts ?
>>>>>
>>>>>I read in the itk book that you can pass the mask to the metric, but as
>>
>>far
>>
>>>>as I
>>>>
>>>>
>>>>>understand it, it helps to improve computational cost ( as the metric is
>>>>>focusing on pixel mask).  But I might misunderstood it...
>>>>>
>>>>>Thanks a lot
>>>>>Delphine
>>>>>_______________________________________________
>>>>>Insight-users mailing list
>>>>>Insight-users at itk.org
>>>>>http://www.itk.org/mailman/listinfo/insight-users
>>>>>
>>>>
>>>
>>>
> 
> 
>