Top-down constraints for level-set based segmentation: shape-priors, symmetry, intensity distributions and user interaction
Speaker: Tammy Riklin Raviv , CSAIL
Date: February 15 2008
Time: 2:00PM to 3:00PM
Location: 32-D507
Host: Polina Golland, CSAIL
Contact: Polina Golland, x38005, polina@csail.mit.edu
Relevant URL: Classical bottom-up segmentation methods, which are based on the image
data alone, are prone to errors in the presence of noise, clutter,
occlusions or shadows. Prior knowledge on the object or the region of
interest can significantly facilitate the segmentation process.
In this talk I will present several segmentation methods, which employ
top-down information together with internal image properties such as
gray-levels and gradients. When a reference object is available, its
shape can be used as a prior. However, accounting for possible
transformations between the different object views, as part of the
segmentation process is challenging. The method suggested accommodates
planar projective transformations by concurrent segmentation and
registration processes. This is accomplished by the construction of a
cost functional, where the dynamic variable is the object boundary
represented by the zero level of a level-set function. The functional
is optimized using calculus of variations.
When only a single image is given but the object taken is known to be
symmetrical, the symmetry provides a significant shape constraint that
can support segmentation. In the second part of my talk I will
present a method to segment objects with either bilateral or
rotational symmetry in the presence of perspective distortion. The
key idea is the use of the symmetrical counterpart image obtained by
flip or rotation of the source image as another view of the object.
In many applications, shape deformations cannot be modeled
parametrically. In these cases other sorts of priors should be sought.
I will present a method for segmentation of anatomical structures in
the mouse brain from histological data. Segmentation is carried out
slice-by-slice where the successful segmentation of one section
provides a prior for the next one. Three-dimensional Gaussian mixtures
are used to model intensities and spatial locations of the region of
interest and the background. This information adaptively propagates
across the sections. I will also demonstrate the delineation of
uterine fibroids in MR images for focused ultrasound treatment. The
algorithm suggested solves actual clinical problem. It is fast and
reliable involving minimal user interaction in the form of few mouse
clicks. Segmentation results are in good comparison with expert
segmentation.
See other events that are part of Biomedical Imaging and Analysis 2007/2008
See other events happening in February 2008
