[Insight-users] Re: a question about the inverse deformation field

[Luis Ibanez]

Hi Lan,

Depending on the characteristics of your computer, that time is
probably reasonable for inverting that size of deformation field.

You may want to check first the amount of RAM memory in your
computer and verify that you are not saturating the memory and
forcing the operating system to swap memory to disk.  You can
check this on Windows by using the task manager, and on Linux
using the top tool, or any other system information tool such
as ksysguard.

Changing the data representation to short may not help much,
and can actually degrade the quality of the output.

Note also that this is an iterative method that runs until
reaching a precision convergence criteria. Depending on your
application, you could relax that convergence criteria and
terminate the filter early on.

Please look at the methods

    SetStopValue( value  )
    GetStopValue()

These method sets the error limit that you should reach in order
to stop the filter. Note that the default value is "ZERO" which
is a VERY demanding precision.  The value is set in millimeters
(or the units of the deformation vectors in your field). So please
choose a value that is consistent with the requirements of your
precision in the problem that you are trying to solve.

You should also look at

     SetNumberOfIterations()


also.... you could add a cout for "smallestError" in line 223
of itkIterativeInverseDeformationFieldImageFilter.txx, that
will help you to see how the convergence progresses at every
iteration.


   Regards,


       Luis



--------------
Lin, Lan wrote:
> Hi Dr Ibanez,
> 
>         I have another question about the inverse deformation field. I 
> got the deformation field from demons registration and used the 
> itkIterativeInverseDeformationFieldImageFilter to inverse the vector 
> field. It took couple of hours to get the output (for example 
> 256*256*168 image registration, the deformation file is 129Mb in float 
> format). Is there anyway to speed up the inverse process? Change the 
> float format to other format such as ushort?
> 
> Thanks in advance
> 
>