[Insight-users] Difficulty registering synthetic volumes with MultiMRIRegistration (fwd)

[Luis Ibanez]

Hi Sarah,

Thanks for the clear description of the problem you are facing.

The reason why the registration fails with your data is that
your .raw files are encoded in pixel type 'int' with 4bytes per
pixel.

the MultiMRIRegistration application is expecting to receive
'unsigned short' as the pixel type of the input.

You have two options:

1) Modify the application's code
2) Reformat your data to be 16bits / pixel


If you decided to go for option (1), it should be enough
to change line # 37 in the file

           MultiResMIRegistration.cxx

in the directory:

    InsightApplications/MultiResMIRegistration/Common

and recompile the application.

The change should be:

    typedef itk::SimpleApp<signed int> AppType;

instead of the current

    typedef itk::SimpleApp<signed short> AppType;


-----

Just for the record:

   Do you really expect to get 32bits/pixel in your
application ?

Most image modalities use 16bits/pixel (e.g. CT) and
seems to be enough for covering the dynamic range.
You may be wasting your disk space and the run-time
memory if you don't really need this precission.

Also, please double check the endianess of your data.
Are you using a Sun or a Mac ?



Regards,


    Luis


---------------------
Sarah E Geneser wrote:
> Hello,
> 
> I am currently trying use the MultiMRIRegistration application to register
> two raw volumes that I have created.  The application seems to parse my
> param file, open and preprocess the images, and begin registration.  The
> problem is that I always get a transform matrix comprised entirely of
> nans.  Also, if I only  use one level of scaling, and one iteration- I
> still get all nans. (This occurs with several learning rate values and
> even with zero iterations specified.)
> 
> I don't think the problem is with my volumes. One volume is a cube
> centered in the image (cube.raw), and the other volume is that same cube
> translated (cube_trans.raw).  I have a tarred and gzipped file of my
> volumes and parameter file (the directories have been changed) at the
> following location:
> http://www.cs.utah.edu/~geneser/cube.tar.gz
> The volumes are 200x200x200 in size and are comprised of unsigned shorts.
> 
> I also have reason to believe that my ITK build is okay. I have gotten
> other applications to work.  If someone could either point me to a pair of
> volumes and corresponding parameter file that is known to work, or suggest
> what might be wrong with my volumes or parameter file - I would really
> appreciate it.
> 
> Thanks,
> 
> Sarah Geneser
> 
>