[Insight-users] Re: On Multiresolution Registration : Pixels vs Millimeters

[Luis Ibanez]

Hi Valli,

I hope any Analyze user out there will
correct me if I'm wrong here.

I would dare to say that the transform
provided by ITK is the right one and
that Analyze is making the omision of not
taking the image spacing into account.

Here is my argument:

1) The SPECT image is about half the number
    of pixels (along X and Y) of the MRI image.

2) The pixel size (or spacing) on the SPECT
    image is about double of the MRI.

This means that both image grids, when mapped
to physical space cover about the same physical
area. For example if we look at one slice:

SPECT =( 120x1.8 , 120x1.8 )=( 216.0 mm, 216mm )
MRI   =( 256x0.86, 256x0.86)=( 220.0 mm, 220mm )

Henceforth, the actual mapping from one the SPECT
image space to the MRI image space should be about
the same scale. That is, diagonal elements close
to "1.0".

I would guess that the anatomical object in the
images fill more or less the volume of both the
SPECT and MRI images. If we disregard the spacing
and assume that both images have the same pixel
size we can rapidly (and wrongly) deduce that the
SPECT image need to be scaled by about 2.0 in order
to make its 120x120 pixels fill the area of the
256x256 pixels of the MRI.

That is, the wrong assumption of spacing = 1.0
SPECT=(120x1.0, 120x1.0) = (120mm, 120mm) (wrong)
MRI  =(256x1.0, 256x1.0) = (256mm, 256mm) (wrong)


This is the unfortunate consequence of thinking
about registration in terms of pixels instead
of millimeters. Transforms in ITK map physical
spaces, not pixel grids. This distintion is quite
important since the size of anatomical features
is meaninful in millimeters, not pixels.

It is useless for a doctor to know that his/her
patient has a tumor of 25.6 pixels in diameter...

ITK is consistent about transforms, registration
and resampling.

If you take the transform resulting from the
registration process and feed it into a Resample
image filter, then connect as input the moving
image used in the registration you would get at
the output the correct mapping of SPECT onto MRI.
Please make sure that you set the following
parameters of the ResampleImageFilter:

- OutputOrigin  to the MRI origin
- OutputSpacing to the MRI spacing
- Size          to the MRI LargestPossibleRegion Size

Note that ITK also uses the image origin when it
comes to register and resample images.

You may want to take a look at the ResampleImageFilter
examples under

        Insight/Examples/Filtering

and the associated documentation on the PDF document

   InsightDocuments/Web/HTML/SoftwareGuide.pdf


Please let us know if you have further questions.


   Thanks


     Luis


-------------------------------

valli gummadi wrote:
> 
> Dear Mr.Luis ,
>          My requirement is to register two datasets taken from different 
> modalities.One is from MRI and another is spect.I have used the 
> multiresolution framework found in itk applications. I got diagnal 
> elements around 1.But,In some other tool called Analyze diagnal elements 
> are around 2.In itk i have to adjust input parameters like translate 
> scale,iterations etc.I have tried adjusting these parameters.But,I never 
> got 2s in the diagnal.I am listing the header details of two data sets.
> 
> MR: 256X256X120 spacing : 0.859373\0.859375\1.60000
> SPECT : 120X120X48 spacing : 1.802125\1.802125\3.60425
> 
> I want to use transformed spect for fusing spect and mri.Because iam 
> failing in registration i could not get the expected output for fusion.
> Please suggest where i am missing actualy?What should i do to get 
> correct output.
> 
> Thanking U,
> 
> Regards,
> Valli
> 
> 
> 
> 
>