[Insight-users] little double about the output of Geodesic Active Contours Segmentation with Shape Guidance

[Haiyong Xu]

Hi Juan,

I assumed you are using
itk::GeodesicActiveContourShapePriorLevelSetImageFilter< >. In brief,
the transform parameters given in output are to define a
itk::Transform object (translation and rotation) that will transform
the target shape to the source shape.

To use this filter, it's really important to understand how
itk::Transform works. This object transforms a point from target to
source, very important, ** not** from source to target. Thinking in
the feature image coordinate system, the source is the original pose
of shape model, which is located at the coordinate origin (0,0) and
rotated by 0 radius. The target is the final pose of shape model,
which is probably located at some position, say (10, 15), and rotated
by pi/4 radius, clockwise. In this example, the output transform
parameters from the ITK program would be: translation (-10, -15),
rotation (-pi/4). In your case, you may need inverse the transform in
order to register the final shape with the image.

There is probably a bug in itk's GeodesicActiveContour segmentation
with Shape Prior framework. I think it mess up with the order of
rotation/translation in itk::Transform object, and still waiting for
the confirmation from ITK developer community. That will cause an
invalid segmentation output. Please see the bug report in
http://www.itk.org/Bug/view.php?id=10617.

Regards,
Haiyong