[Insight-developers] Re: MR Segmentation

[Stephen R. Aylward]

Hi Celina,

Thanks for the suggestions.

Stephen

Celina Imielinska wrote:
> 
>  Stephen,
> 
>    why do you think that to segment "everything in the pelvic area"
> "wathershed is a great approach for this"? Have you seen validation
> of wathershed, or any other segmentation method (under itk) for the pelvic
> region? Is it a male pelvic or female pelvic region? If these are
> "all structures" then it must also include the all fine muscles in
> the perineum that anatomists still argue how to identify.
> Do you know how to generate ground truth for the structures in the
> region? I don't, and I know that even anatomists don't know either.
> From our experience with segmentation of the Visible Human male pelvis
> (and we have on our team an anatomist who teaches about the region), this
> is one of the most challenging regions to segment.
> 
>   I think that before segmentation methods developed under itk will
> be systematically validated, we can't make statements about what method
> is the the best for what application (what imaging). We really can't make
> conjectures before we will validate...
> 
>  -Celina
> 
> 
> 
> On Tue, 8 Jan 2002, Stephen R. Aylward wrote:
> 
> > Hi Nils,
> >
> > I think I misunderstood your problem - you don't want just the bones -
> > you want everything in the pelvic area. Right?  Sorry for the confusion.
> >
> > Watershed is a great approach for this.   Fuzzy connectedness should
> > also be considered.   It is a hard problem, though.   This is one
> > application that really needs a user interface/visualization in order to
> > evaluate, control, and edit the segmentation process.   Stay tuned for
> > itk2 :)
> >
> > As you go along, if you have suggestions for how we can better group the
> > methods, summarize their utility, etc, we would really appreciate it.
> > One thing, this year is going to focus on documenting the application of
> > itk to a variety of different medical image analysis tasks (as part of
> > method validation), and so the examples section is going to greatly
> > expand and perhaps that will help others in your position in the future.
> > Until then....
> >
> > Thanks for your patience,
> > Stephen
> >
> > Nils Hanssen wrote:
> > >
> > > Hi Stephen,
> > >
> > > thank you very much for your step-by-step list. In fact i just had a
> > > conversation with lydia ycl about segmentation in the pelvic area because
> > > she deals with this topic.
> > >
> > > Instead of being limited to the pelvic area, we have a general interest in
> > > the segmentation capabilities of itk for CT and MR data. So far, we used vtk
> > > for visualization and used thresholding as well as region growing for
> > > segmentation and marching cubes for surface generation of bony structures.
> > > Since vtk's segmentation capabilities are (obviously) very limited, we were
> > > looking forward to become a first version of itk.
> > >
> > > I am very impressed by itk from the software-engineering point of view! I
> > > like the concepts and the realization of some design patterns. However, it's
> > > not easy to get an overview of the whole system and how everything belongs
> > > together.
> > >
> > > Currently i try to get the streamed watershed filter running.
> > >
> > > Nils
> > >
> > > > It looks like Nils wants to segment the pelvis from CT and MR.
> > > >
> > > > This should be do-able with itk now...
> > > >
> > > > 1) Region grow (itkRegionGrowImageFilter or
> > > > itkFuzzyConnectednessImageFilter.h)
> > > > 2) morphological erosion (2cm - remove ribs)
> > > > 3) morphological dilate (3cm - restore to original size +
> > > > fill in marrow
> > > > holes)
> > > > 4) morphological erosion (1cm - restore to original size)
> > > >
> > > > The problem with abdominal MR are the intensity inhomogeneities.
> > > >
> > > > Try the MRBiasFieldCorrection method and then do the above
> > > > for MR (bones
> > > > being dark).
> > > >
> > > > Stephen
> > > >
> > > >
> > > >
> > > > --
> > > > ===============================================
> > > > Dr. Stephen R. Aylward
> > > > Assistant Professor of Radiology
> > > > Adjunct Assistant Professor of Computer Science
> > > > http://caddlab.rad.unc.edu
> > > > aylward@unc.edu
> > > > (919) 966-9695
> > > >
> >
> > --
> > ===============================================
> > Dr. Stephen R. Aylward
> > Assistant Professor of Radiology
> > Adjunct Assistant Professor of Computer Science
> > http://caddlab.rad.unc.edu
> > aylward@unc.edu
> > (919) 966-9695
> > _______________________________________________
> > Insight-developers mailing list
> > Insight-developers@public.kitware.com
> > http://public.kitware.com/mailman/listinfo/insight-developers
> >

-- 
===============================================
Dr. Stephen R. Aylward
Assistant Professor of Radiology
Adjunct Assistant Professor of Computer Science
http://caddlab.rad.unc.edu
aylward@unc.edu
(919) 966-9695