[Insight-users] Re: software giude

[Luis Ibanez]

Hi Natasha,


Thanks for pointing this out.


You are right,

              The heuristics suggested for the
alpha parameter in the Sigmoid are misleading.


The computation for alpha was supposed to
produce a negative number. This will result
in the intensity inversion that we need for
converting bright contours from the gradient
magnitude into dark contours for the speed
image. The current suggested value need an
extra minus sign, or as you propose, K1 and
K2 should be reversed.



The suggested value of alpha is now described as

        alpha = ( K2 - K1 ) / 6

with the explicit statement that K2 is expected
to be lower than K1 and therefore we expect
alpha to be negative.

This has been fixed in the embedded text in the
example  FastMarchingImageFilter.cxx and will
show up in the PDF document for the next update
of the guide.



Please let us know if you find any other
errors in the documentation.


   Regards,


     Luis




------------------------
Natasha Kovacevic wrote:

> Dear Luis, 
> 
> I tried using fast marching segmentation on 3D images. After playing
> with various parameters I came to the conclusion that Software Guide
> might have a typo on page 369. Suggested values for sigmoid alpha and
> beta are (K1-K2)/6 and (K1+K2)/2 respectively. Presently, K1 is defined
> as the minimum gradient magnitude along the edge of the structure, while
> K2 is defined to be the average gradient magnitude within the interior
> of the structure. I believe that you want to switch the definitions of
> K1 and K2. At least that makes sense for the speed image and certainly
> gave good results with my data.
> 
> Regards, Natasha
> 
> Natasa Kovacevic, Ph.D
> Mouse Imaging Centre
> Hospital For Sick Children
> 555 University Avenue
> Toronto, ON M5G 1X8
> Canada
>   
> tel. (416) 586-4800/6511
> nkovacev at phenogenomics.ca
> 
> 
> 
>