[Insight-users] Suggestion(s) for rigid registration

[Luis Ibanez]

Hi Martin,

You are dealing with quite a difficult registration problem.


About you questions:

1) Correcting the signal bias will help. However the bias
    distribution that you have in the floating image is not
    the typical bias present in MRI. You actually have high
    intensities in two regions close to the borders plus
    the image seem to be cropped in a non-orthogonal field
    of view (the elliptical mask that you mention).

    You will have to modify the MRIBiasField correction filter
    in order to consider *only* the pixels inside the elliptical
    mask. You probably can do this too by inserting a threshold
    criteria in the pixels values being considered by the MRI
    Bias field corrections filter.

2) It is difficult to say what will be an *optimal* metric
    for this case... but given that it it a multi-modality
    case, you will have to choose between any of the Mutual
    Information implementation available in ITK.

    You probably want to try first the implementation of
    the Mattes Mutual Information. Note that not all the
    Image Metrics in ITK support the masking option.
    MattesMI will support it. You can use an itkSpatialObject
    in order to provide the elliptical mask
http://www.itk.org/Insight/Doxygen/html/classitk_1_1EllipseSpatialObject.html

    You will find examples on the use of Masking on the directory

         Insight/Examples/Registration



3) If you start using Mutual Information metrics, you
    probably want to try the OnePlusOne evolutionary
    algorithm. This algorithm is robust for optimizing
    noisy functions.



  Regards,


     Luis




--------------------
Martin Kavec wrote:

> Hi,
> 
> I am rather new to image registration and would like to hear your opinion
> to solve the problem the most efficiently. Before I would start with ITK
> implementation, I tried to rigid register the images using FSL or
> Analyze6, but these failed.
> 
> I would very much appreciate if you have a look at the volumes. You can
> grab them here :
> 
> http://www.uku.fi/~kavec/volumes.tar.bz2
> 
> The volumes are in Analyze format tar-ed and bz2-ed. The images are 3D
> MRI volumes of the subject scanned on a different field strengths. The
> reference.* is a high-field T1-weighted and I suppose its fine. However,
> floating.* is a low-field "T1-weighted", lower resolution. In addition,
> large surface coil was used (note high signal on the top of the skull and
> nose), plus the image is by default (by manufacturer) masked using
> elliptical mask.
> 
> The problems I can see, which cause a registration failure are large
> signal intensities variation due to surface coil and the elliptical mask.
> 
> Questions:
> 
> 1. Correcting the signal bias (using itkMRIBiasFieldCorrectionFilter)
>    would increase the chances for better registration outcome?
> 
> 2. Normalized mutual information would be optimal cost function, taking
>    into account elliptical mask?
> 
> 3. Any other suggestions (ITK optimizer, ...)?
> 
> 
> Thanks for help in advance.
> 
> Martin
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 
>