[Insight-users] VectorImage Registration Metric

[Luis Ibanez]

Hi Lisa,

You are right,
Thanks for pointing this out.


Yes, we should move the code into the Sandbox.
(please check first if Jeffrey has defined any License
for his version of the code, if so, we have to verify if
that the license is BSD compatible, before we adopt
the code).


Do you have a NAMICSandbox  write account ?

If not, please request one at:
https://www.kitware.com/Admin/SendPassword.cgi


Also, about your question:

Yes, if you want to extend registration to tensor images,
a similar extension will have to be made. You will also
need a TensorResampleFilter based on the itkVector
ResampleFilter.

Please note however that Interpolations for Tensors
may require different strategies, depending on what
the Tensor is representing.  For example, when the
Tensor represents some physical property, it may
be subject to some constrains.

In the NAMIC group there has been some work related
to Tensor interpolation.

Regarding the actual Metric, the adaptation of the
GetValue() method is rather easy. You simply
need to introduce a measure of distance between
two Tensors (again, probably based on what the
Tensors are representing).

The more involved piece of work is the computation
of derivatives.

The Derivative of the metric will involve the computation
of a third-rank Tensor, which is the result of deriving
the image Tensors with respect to space.

This could be represented in 3D matrix of N components,
where N is the dimension of the image Space.

It will be nice to develop this code also in a NAMICSandbox
directory, so you can benefit from the feedback of
other groups in the community.


Let us know if you need any help,
we will be happy to assist you.


         Luis



-----------------------------------------------------------------------
On Tue, Feb 9, 2010 at 2:21 PM, Lisa Tang <lisat at sfu.ca> wrote:
> Hello Luis,
>
> Thanks for your reply.
>
> On closer look, I think these items have already been fixed by Jeffrey Dudas, whose work was discussed in Oct 2009
> http://old.nabble.com/Question-about-multiResolutionPyramidImageFilter-class-to25839179.html#a25862138
>
> For example, the code for gradient calculation has been added here:
> http://www.seas.upenn.edu/~jtduda/code/itkVectorImageToImageMetric.txx
> http://www.seas.upenn.edu/~jtduda/code/itkVectorMeanSquaresImageToImageMetric.txx
>
> Perhaps the sandbox needs to be updated with his work?
>
> A related question is its extension to Tensor images. A needed class/filter that comes to mind is for the purpose of reorienting tensors after the tensor image is transformed. Could you please point us to how we can proceed with this?
>
> e.g. should it follow construct of itkVectorResampleImageFilter, but using itkDiffusionTensor3D as input?
>
> Thanks again,
> Lisa
>
>
> ----- "Luis Ibanez" <luis.ibanez at kitware.com> wrote:
>
>> From: "Luis Ibanez" <luis.ibanez at kitware.com>
>> To: "Lisa Tang" <lisat at sfu.ca>
>> Cc: "Ghassan Hamarneh" <hamarneh at cs.sfu.ca>, "itk" <Insight-users at itk.org>
>> Sent: Tuesday, 9 February, 2010 10:46:19 GMT -08:00 US/Canada Pacific
>> Subject: Re: VectorImage Registration Metric
>>
>> Hi Lisa,
>>
>> Please look at the file:
>>
>> NAMICSandBox/VectorImageRegistrationMethod/Source/
>>                     itkVectorMeanSquaresImageToImageMetric.txx
>>
>> and the lines that have "FIXME" messages as comments.
>>
>>
>> These are the lines where a change from "Scalar", to "Vector"
>> computation remains to be introduced.
>>
>> For example, the "gradient", used to be a Vector in the
>> computation of a Metric for a Scalar image, but when
>> doing that computation for  a Vector image, the gradient
>> must be a Matrix, which is composed of the gradients
>> of each individual component of the vector image.
>>
>>
>> Please let us know if you see any problems with
>> the suggested implementation,
>>
>>
>>       Thanks
>>
>>
>>             Luis
>>
>>
>> ------------------------------------------------------------------------------------------
>> On Fri, Feb 5, 2010 at 2:30 PM, Lisa Tang <lisat at sfu.ca> wrote:
>> > Hi Luis,
>> >
>> > You mentioned that the version in NAMIC sandbox need "massaging":
>> >
>> > "The Gradient of the vector image must be computed as a matrix
>> > and multiplied with the Transform jacobian in order to produce
>> > another matrix, that then will be collapsed into a gradient
>> > vector. (to be done)"
>> >
>> > However, I find that both the Sandbox's
>> itkVectorMeanSquaresImageToImageMetric and
>> itkMeanSquaresImageToImageMetric calculate the metric's derivative
>> similarly, and so are the calculations of the gradient image in
>> itkVectormageToImageMetric.
>> >
>> > Could you please clarify which class were you referring to that
>> requires the above change?
>> >
>> > Thanks in advance,
>> > Lisa
>> >
>> >
>> >
>> >
>> > ----- "Luis Ibanez" <luis.ibanez at kitware.com> wrote:
>> >
>> >> From: "Luis Ibanez" <luis.ibanez at kitware.com>
>> >> To: "Ghassan Hamarneh" <hamarneh at cs.sfu.ca>
>> >> Cc: "Lisa Tang" <lisat at sfu.ca>
>> >> Sent: Friday, 5 February, 2010 07:04:30 GMT -08:00 US/Canada
>> Pacific
>> >> Subject: Re: VectorImage Registration Metric
>> >>
>> >> Hi Ghassan,
>> >>
>> >> I haven't seen much activity in this project for a while,
>> >> but sometimes research groups make progress without
>> >> necessarily sharing their day-to-day updates with the
>> >> community.
>> >>
>> >> I would suggest to bring up the topic to the ITK users
>> >> and developers list.
>> >>
>> >> I will be happy to ask the question in the lists if you
>> >> prefer me to do it.
>> >>
>> >>
>> >>      Regards,
>> >>
>> >>
>> >>           Luis
>> >>
>> >>
>> >>
>> >> ---------------------------------------------------------
>> >> On Fri, Feb 5, 2010 at 12:55 AM, Ghassan Hamarneh
>> <hamarneh at cs.sfu.ca>
>> >> wrote:
>> >> >
>> >> > Hello Luis
>> >> >
>> >> > Hope all is well!
>> >> >
>> >> > I came across this thread:
>> >>  http://www.cmake.org/pipermail/insight-users/2008-May/025746.html
>> >> >
>> >> > We were thinking of contributing something similar (vector
>> valued
>> >> image registration) to the insight journal as was developed for
>> our
>> >> paper:
>> >> >
>> >> > [ SMRFI: Shape Matching via ] Registration of Vector-Valued
>> Feature
>> >> Images
>> >> > http://www.cs.sfu.ca/~hamarneh/ecopy/cvpr2008.pdf
>> >> >
>> >> > Before investing more time on this, we appreciate if you could
>> let
>> >> us know the status of this effort, is any one working on this,
>> etc?
>> >> >
>> >> > Thanks,
>> >> >
>> >> > /Ghassan
>> >> >
>> >> >
>> >> >
>> >> >
>> >> >
>> >
>