[Insight-users] Segmentation

[Sayan Pathak]

Hi Suresh,
In addition to what Luis mentioned. There is also a classifier framework =
that is suitable for this purpose. It does a great job on gray and white =
matter. This framework also supports multivariate analysis. So if you =
have multiple channels of data, this framework could be quite useful. =
You might consider looking into it. You can find some test codes in ITK =
algorithms tests itkSupervisedImageClassifierTest, itkMRFImageFilterTest =
etc. Please feel free to contact us for more further assistance.

Sayan



> -----Original Message-----
> From: Luis Ibanez [mailto:luis.ibanez@kitware.com]
> Sent: Friday, August 23, 2002 6:31 AM
> To: suresh
> Cc: insight-users@public.kitware.com
> Subject: Re: [Insight-users] Segmentation
>=20
>=20
>=20
> Hi Suresh,
>=20
> If you don't mind to do a bit of user interaction,
> an easy way of getting gray matter, white matter
> and csf is to use a pipeline similar to the one
> used in the RegionGrowingSegmentation example:
>=20
> http://www.itk.org/HTML/RegionGrowingSegmentation.htm
>=20
> Which is basically:
>=20
> 1) Use one edge-preserving smoothing filter,
>     for example:
>=20
>     - CurvatureFlow
>     - GradientAnisotropicDiffusion
>     - CurvatureAnisotropicDiffusion
>=20
> 2) Use the ConfidenceConnectedImageFilter.
>     Which requires three parameters:
>=20
>     a) a multiplier for the variance of the
>        gray level distribution of the tissue
>        type you are interested on.
>=20
>     b) a number of iterations
>=20
>     c) a seed point lying on the tissue of
>        interest.
>=20
> (a) can be fixed at 2.5 for most purpouses
>=20
> (b) can also be fixed to something around 5
>=20
> (c) will be better if provided by a user
>      clicking on the image.
>      This is very easy for a user and will
>      relieve your application from dealing
>      with computer vision issues.
>=20
>=20
> Combining effective user-interaction with
> supervised methods is probably the best
> trade-off in medical imaging today.
>=20
> --
>=20
> If you are looking to process a large number
> of images and consider user-interaction to
> be undesirable, a good alternative could be
> to rely on a rough registration of your data
> with a presegmented brain just with the aim
> of obtaining three standard seed points.
>=20
> A MultiResolution Rigid registration using
> a metric like MeanSquare should converge
> rapidly for this kind of application.
>=20
> Please let us know if you will like to have
> more details.
>=20
>=20
>    Thanks
>=20
>=20
>     Luis
>=20
>=20
> =
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>=20
> suresh wrote:
> > Hi Luis,
> >=20
> > Thankyou for your elaborate reply.
> > I'm not interested in detailed brain structure.Just=20
> separating { White=20
> > matter, Gray matter, CSF }
> > will be fair enough.
> >=20
> > Can you please help me with a little more details.
> >=20
> > Thank you.
> >=20
> > suresh
> >=20
> >=20
> > _______________________________________________
> > Insight-users mailing list
> > Insight-users@public.kitware.com
> > http://public.kitware.com/mailman/listinfo/insight-users
> >=20
>=20
>=20
>=20
>=20
> _______________________________________________
> Insight-users mailing list
> Insight-users@public.kitware.com
> http://public.kitware.com/mailman/listinfo/insight-users
>=20
>=20