[Insight-users] Do RecursiveHessian or RecursiveGaussian require isotropic voxels?

[Luis Ibanez]

Hi Jerome,

No, the RecursiveGaussian filter does not requires the voxels
to be isotropic.

The RecursiveGaussian filter takes the Pixel size into account.

This means that the computations are performed correctly in
physical units.

The values reported by the RecursiveGaussian and its associated
filters must be interpreted in terms of derivatives with respect
to millimeters.

Note that if what you are doing is multi-scale analysis (and it
sounds like that, since you are searching for the sigma that matches
a particular blob...) then you should use the Normalize along scales
option, which will make possible for you to compare the results from
one scale (sigma) with the results of a different scale (sigma).


---

Just for the record...

What is the size of the Blobs in your experiments in pixels ?
and what is the size of those pixels in millimeters ?


If you find that the behavior of the filter is incorrect, please
post a minimal piece of code that illustrates your point, and if
necessary post an input/output image on a web accessible place.


Best Regards,


    Luis



----------------------
SCHMID, Jerome wrote:
> Dear Itk users,
> 
> I am playing with vesselness measure based on eigen values of Hessian 
> matrices. I have noticed a quite odd behaviour and I would like to know 
> if it is normal.
> 
> Basically the eigenvalues change "a lot" whether or not my input image 
> has or not isotropic voxels. The "a lot" changing is related to the 
> tubes or blobs detection with the Hessian+eigenValues principle, i.e. 
> roughly lambda1 = lambda2 = lambda3 for a blob and lambda1 = 0 and 
> lambda1 << 0 and lambda3 << 0 for a tube.
> 
> In my case with a blob and non isotropic image: lambda1 close to 0 and 
> lambda2/3 << 0. -> odd(?)
> In my case with a blob and isotropic image: lambda1 = lambda2 = lambda3 
> (roughly) -> expected
> 
> Knowing that the hessian is computed with the 
> RecursiveGaussianImageFilter that is based on the RecursiveGaussian 
> filter, I wonder whether the pb is here.
> 
> A note: my sigma for the detection is close to the radius of the blob, 
> and it is defined in physical units. I expect that the recursive 
> gaussian filter uses sigma in physical units ( which looks to be the 
> case, looking at the distinction variance/sigma, isn'it?).
> 
> Any suggestion? Thanks!
> 
> Best Regards,
> 
> Jerome Schmid
> 
> -----------------------------------
> Jerome SCHMID
> Project Manager/ Engineer
> Augmented and Virtual Reality
> MIS Centre
> Prince of Wales Hospital
> Chinese University Of Hong-Kong
> -----------------------------------
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users