[Insight-users] Re: RG_Active contour Parameters : and how to post to the Insight Journal

[Luis Ibanez]

Hi Yasser,

1) Start with only 1 iteration
2) Start with a multiplier of 1.5
3) Use a neighborhood size of 2
4) Use as many seed points as you
    can place in the homogeneous
    region that you are segmenting


Your first segmentation will be plenty
of holes and missing pieces because
a multiplier of 1.5 is too restrictive.
What is important in your first run
is to get *only* pieces from the region
that you want to segment. That is, make
sure that you don't get any section that
does not belong to the object that you
want to segment.

Once you get *only* pieces from the desired
object, you can rerun from the start using
Multipliers that are larger. The value of
2.5 is typically a good compromise, while
a value of 3.0 is usually too large. This
of course is based on typical images with
3% of Gaussian noise.  If your images are
noiser you will need larger multipliers,
if you images have less noise you will need
smaller multipliers.

Once you find a good multiplier value, you
can play with providing less and less
seed points.  You may be able to get as
low as one or two seed points, assuming
that you place them in locations that are
very representative of the object.

Once you reduce the number of seed points
you can experiment by running the algorithm
for multiple iterations. Typically you dont
want to go further than 5 iterations. Note
that the behavior of this algorithm is not
obviously intuitive with respect to the number
of iterations.

More iteration may damage a currently good
segmentation.

If you still have many remaining holes, instead
of running more iterations you may want to use
the HoleFilling Voting filters that are described
in the ITK Software Guide.


    http://www.itk.org/ItkSoftwareGuide.pdf


BTW, If you find good parameters for segmenting
the IBSR images, you should submit a paper to
the Insight Journal, describing the parameters
that you used and the image that you obtained.

The Insight Journal will provide the mechanisms
for future users to take advantaqe of the experience
that you are going to gain with your current efforts.

You will find the Insight Journal at:

http://www.insightsoftwareconsortium.org/InsightJournal/

Note that the Insight Journal is not the place for
the typical "My-method-is-better-than-yours" kind
of papers. The Insigh Journal is based on the concept
of reproducibility that is fundamental for serious
scientific work. A paper to the Insight Journal must
be an account on how to repeat the work of the authors.
"Original work", "Innovation" and those other mortal
vanities are not required in the Insight Journal, because
after all, they are not part of the scientific method.



Regards,



    Luis




-------------------
yasser salman wrote:
> Hi every one , hi luis
> i'm using IBSR data ("Another subject with tumor,
> multiple scans over time" series 32i, 45i,....88i) to
> segment tumor volume using confidance connected and
> active contour segmentation , can u tell me the
> suitable parameters to start the segmention process
> for both methods, in these series,
> thanks,
> yasser ,
> 
> 
> 		
> __________________________________ 
> Yahoo! Mail for Mobile 
> Take Yahoo! Mail with you! Check email on your mobile phone. 
> http://mobile.yahoo.com/learn/mail 
> 
>