[Insight-users] Segmentation Question : Breast MRI

[Luis Ibanez]

Hi Neha,

The choice of a good candidate method for detecting
breast abnormalities depends on the characteristics
of the Breast MRI image you have.

Do you have pairs of images before and after
injection of gadolinium contrast agent ?

If so, would you need deformable registration
between the pair, or are the images acquired
in well controled conditions ?

If you dispose of well aligned pairs, one good
option to start with is subtracting the images and
applying Watersheds on the resulting difference.

You may want to experiment with the segmentation
editor application available in InsightApplications.
There is a recent version using an FLTK GUI interface.

Another option is the set of Geodesic Morphology filters

http://www . itk . org/Insight/Doxygen/html/classitk_1_1GrayscaleGeodesicDilateImageFilter . html
http://www . itk . org/Insight/Doxygen/html/classitk_1_1GrayscaleGeodesicErodeImageFilter . html

and the FillHoleImage filter (based on the previous ones)

http://www . itk . org/Insight/Doxygen/html/classitk_1_1GrayscaleFillholeImageFilter . html

to be applied on the result of subtracting the two
MRI breast image (pre and post gadolinium injection).


Once you have identified features in the images, you could
use the ConnectedComponents filter for computing statistics
on the size of each feature.

After that, you will have to consider morphological
criteria for rejecting/accepting features as abnormalities.

----

You also will have to evaluate the noise level of the
image since it will determine a limitation on the size of
the smallest abnormality that could be detected.

-----

If the images happen to be missaligned,... you better
start gathering coffee suplies and take a look at the
mutual information metrics and the BSpline deformable
transform.



    Regards,


       Luis


---------------
Neha D wrote:
> Hello All,
>  
> I plan to use ITK to segment different regions of Breat MRI.
> Does ITK facilitate the delineation/quantification of abnormailities?
>  
> What algorithm would be best to use on these images to find out if the 
> image contains the abnormaility or it is a normal breast image.
>  
> Thanks.
>