[Insight-users] segmentation problems using Fast marching image filter

[Dan Mueller]

Hi Vijay,

The result of the FastMarchingImageFilter is heavily dependent on the 
speed image (which is provided as the filter input): if the speed image 
does not contain the features you require, the results will not be 
meaningful. Further, because the contour is controlled by position only 
(ie. the front expands outwards according to the speed image values - 
not curvature or advection), you must ensure that the important edges of 
your speed image are close to 1.0 to ensure the front does not 'leak' 
into unwanted regions too early.

I highly recommend that you read the section on Fast Marching 
segmentation in the Insight Software Guide ( 
http://www.itk.org/ItkSoftwareGuide.pdf , around section 9.3.1) and also 
(if you can lay your hands on it) the book by Sethian mentioned in the 
Algorithms/itkFastMarchingImageFilter.h header file/documentation. After 
doing this reading, it is probably worthwhile to comment out most of 
your Fast Marching code (leaving only the speed image computation part) 
and write out the speed image to disk to ensure it captures the 
edges/features you are interested in. Also, be mindful of your image 
viewer when viewing the output of the Fast Marching method. In most 
image viewers, the output will typically look binary (black and white 
only) because the parts of the image the front did not reach will be 
assigned a large value (in the order of 3.402823e38) whereas the parts 
the front did reach will be assigned the time-crossing value (eg. 
0-1000) -- 1000 compared to 3.402823e38 is very small and will appear 
black. You may want to threshold the output between [0 .. StoppingValue] 
to strip away these very large portions of the output.

In regards to supplying your image to the ITK community: it is probably 
best to upload the image to some web-based file share system and paste 
the link into an email to the insight-users list.

HTH

Cheers, Dan

vijay kumar wrote:
> Hi,
> I modified FastMarchingImageFilter.cxx that is found in 
> Examples\Segmentation to work on a 3D analyze format image. After 
> compilation, I executed it from the command line:
>  
> FM_Filter3D.exe mri4itk.hdr mri4itkResult.hdr
> 1.25 -8.0 75.0 10 15
> 108 125 1 110 125 5 112 124 10 113 124 15 113 123 20 107 130 25 94 136 32
>  
> The second line has values for sigma, alpha, beta, threshold and 
> stopping time
> The third line contains indices (x,y,z) for a set of 7 points. All 
> three lines were concatenated into one when running the program. The 
> segmentation results don't look good. The binary image contains a 
> large white spot in each slice. I am trying to segment the right 
> artery which appears as a white spot on the right side in the input 
> image. The text of my program is given below. I tried to attach the 
> image (mri4itk.hdr, mri4itk.img) to this email, but  the email 
> bounced. The size is approximately 3 MB. Can I send the images somehow?
>  
> Thanks,
> Vijay
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20070501/4702381b/attachment.html