[Insight-users] Post-process Hessian Sheet Analysis

[paniwani]

Hi,

I'm interested in detecting submerged folds and polyps in virtual
colonoscopy CT images. I've studied the various Hessian code in ITK
including Frangi and Sato, and developed a formula to detect dark sheets
(dark folds surrounded by stool), but my final images still contain a lot of
noise. I need some suggestions on how to best post process the hessian
response. 

Dark sheets should be characterized by 1 large positive eigenvalue (Lambda3)
and 2 eigenvalues close to 0 (Lambda1 and Lambda2). As such, I used the
following scheme:

I ordered the eigenvalues by absolute magnitude as follows:

|Lambda1| < |Lambda2| < |Lambda3|

I only search in tagged regions, i.e. where the intensity is greater than
150 HU.

In these regions, I apply the following enhancement function to detect
folds:

if ( Lambda3 > 0)

H = abs( Lambda3 ) * ( 1 - abs(Lambda1/Lambda3) )^gamma * (1 -
abs(Lambda2/Lambda3) )^gamma

I apply this formula in a multi-scale approach, taking the highest output
across two sigma scales (0.56 and 1.4 mm) with a gamma of 3.

My current output (attached) is very noisy. How can I best reduce this noise
and only enhance the folds? 

Some of the current suggestions I have are to:

1) Threshold the output
2) Remove components smaller than a certain size
3) Remove components not connected to tissue

However, the problem with this approach is that after thresholding, some
components become disconnnected and are then removed after the connected
component filter. If I reduce the threshold or remove it entirely, erroneous
components from stool become attached in the connected components and are
not removed. Are there other, more general, approaches to post processing
hessian output?
http://itk-insight-users.2283740.n2.nabble.com/file/n6234541/Modified_mr10_092_13p.i0344_85_100_input_isotropic.jpg 
http://itk-insight-users.2283740.n2.nabble.com/file/n6234541/Modified_mr10_092_13p.i0344_85_100_hessian.jpg 

--
View this message in context: http://itk-insight-users.2283740.n2.nabble.com/Post-process-Hessian-Sheet-Analysis-tp6234541p6234541.html
Sent from the ITK Insight Users mailing list archive at Nabble.com.