[Insight-users] Need help regarding registration techniques (ITK newbie :) )

[BalajiR]

Hello Sven and Karthik.

Thanks for your suggestions.
After studying chapter 8 of the ITK software guide, I realized there are
many parameters that I
need to optimize to suit my needs. 

I will try these out and let you know the results.

Here are some of the details of my implementation. 
- CT image size 512 x 512 and PET image size 128 x 128
  CT has higher resolution and currently the B-Spline grid is defined
keeping CT as the fixed image.
  I will try with PET as the fixed image.

- B-Spline grid dimensions (8 x 8). I will try modifying this parameter.

- I will look into the multi-resolution options.

- I have tried caching but the program crashes because of lack of memory.

- Using localized optimization is not suitable since the region is not
known.

- Rigid transformation followed by Deformable is also not suitable since the
amount of rigid transformation  is also not known.

Thanks again for your suggestions.

Regards,
Balaji




Karthik Krishnan-2 wrote:
> 
> On Thu, Aug 20, 2009 at 4:28 AM, Balaji, R IN BLR SISL
> <r.balaji at siemens.com
>> wrote:
> 
>> Can you please advice on which algo will be suitable and if possible the
>> reason behind the time taken by the B-Spline algo ?
>>
> 
> Balaji:
> 
> Yes it is quite time consuming. The following questions will help
> ameliorate
> your situation.
> 
> - What is the resolution of the images ?
> 
> - On which image is the BSpline grid being defined PET or CT. Choosing the
> lower resolution image as the fixed image is going to speed up metric
> evaluation.
> 
> - What is the resolution of the grid ? This impacts speed.
> 
> - Have you turned ITK_USE_OPTIMIZED_REGISTRATION on ?
> 
> - If the answer to the previous question is yes, did you enable caching of
> the BSpline transform parameters ? This provides significant speedups. See
> http://www.insight-journal.org/browse/publication/172 for quantitative
> experiments detailing the speedups.
>   Depending on the resolution of your data and memory availability on your
> machine, this may or may not be possible. Memory requirements shoot up
> with
> caching.
> 
> - Did you take a look at multiresolution registration options. Take a look
> at DeformableRegistration15.cxx . This does a coarse and fine deformable
> registration and uses a BSplineDecomposition filter to initialize the fine
> BSpline transform from the coarse.
> 
> - Did you take a look at rigid followed by deformable registration. Again
> please look at DeformableRegistration15.cxx .
> 
> - If your registration problem can be localized to a specific region, you
> could specify a fixed image region and/or fixed and moving image masks.
> This
> will restrict the computation of the metric / and / allow you to define a
> higher resolution transform within the region in question.
> 
> Hope this helps.
> 
> 
> 
> -- 
> Karthik Krishnan
> R&D Engineer,
> Kitware Inc.
> Ph: 518 881 4919
> Fax: 518 371 4573
> 
> _____________________________________
> Powered by www.kitware.com
> 
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
> 
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
> 
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
> 
> 

-- 
View this message in context: http://www.nabble.com/Need-help-regarding-registration-techniques-%28ITK-newbie-%3A%29-%29-tp25057902p25074692.html
Sent from the ITK - Users mailing list archive at Nabble.com.