[Insight-users] Parameters to fine tune

[Luis Ibanez]

Hi Yasser,


             The short answer is:     No   :-)


There are no algorithms without numeric parameters.


          "Automatic" segmentation doesn't exist.


It is an urban legend that has only been possible thanks
to the fact that most papers published in Journals are not
replicated by anybody, and that therefore authors can get
away with making claims that are quite "adventurous".


The closest that you can get to "automatic" segmentation,
is by first taking care of the "manual" part of every algorithm.


This "manual" part is indeed the effort for fine tunning the
parameters of the algorithm to make it match the
characteristics of the image collection that you need
to segment


There are however, some helper methods that can assist
you in finding some of such parameters.

For example, if you are looking for a threshold to segment
two regions that have different intensities, you could look
at the Otsu threshold estimation method:
http://www.itk.org/Insight/Doxygen/html/classitk_1_1OtsuThresholdImageCalculator.html


Note however, that this method is based on certain
assumptions that may or may not apply to the images
that your are segmenting.


For example,
have your MRI images been corrected for field
inhomogeneities ?


    Regards,


           Luis


----------------------------------------------------------------------------
On Sat, Feb 27, 2010 at 6:53 AM, yasser salman <yass71 at yahoo.com> wrote:
> Hello all,
>
> I segmented tumor from MRI data-sets in 3D using different techniques (Region growing , level sets , ) but these segmentation techniques  have numerical parameters that you will have to fine tune. so it takes massive time for tuning, can i found any of the automatic segmentation methods that can extract My ROI in brain,  on other hand in threshold Region growing how can i set L/H threshold in automatic way?
> Yasser
>
>
>
>