Biomedical Engineering Reference
In-Depth Information
a method known as “live-wire” with which a user roughly draws the boundary
of a region of interest. An automatic process then takes over and revises the
boundary by optimizing a cost function. An alternative method is introduced
that allows the user to select a number of points on the region boundary, and
the program then automatically finds boundary segments between consecutive
points, again by minimizing the related cost functions. These methods have
been optimized for speed [9]. They have also been extended to 3D [7]. In 3D,
the program receives boundary contours in a few strategically placed slices and
produces contours in other slices.
Cabral
et al.
[4] describe editing tools that are associated with a region-
growing method, enabling a user to add or remove image voxels in a region to
revise the region. Hinshaw and Brinkley [14] developed a 3D shape model that
uses prior knowledge of an object's structure to guide the search for the object.
Object structure is interactively specified with a graphical user interface.
H ohne and Hanson [15] developed low-level segmentation functions based
on morphological operators that interactively delineate regions of interest. Pizer
et al.
[24] describe a method that segments a volumetric image into regions at
a hierarchy of resolutions. Then, the user, by pointing to an object in a cross-
sectional image at a certain resolution, selects and revises a region. Welte
et
al.
[27] describe an interactive method for separating vessels from each other
and from the background in MR angiographic images. To reduce the complexity
of the displayed structures during the interactive segmentation, a capability to
select substructures of interest is provided.
Energy-minimizing models or “snakes” are another set of tools that can be
used to guide a segmentation and revise the obtained results [18, 20]. With an
energy-minimizing model, a contour or a wireframe is initiated approximately
where an object of interest is believed to exist. An optimization process is then
activated to iteratively revise the contour or the wireframe to minimize a local
cost function that defines the energy of the snake. Since some points in a snake
may trap in local minima, the globally optimal solution may be missed. To avoid
this, often the user is allowed to intervene and either move some of the snake's
points that are thought to have converged to local minima, or guide the snake
to the optimal position by interactively controlling the external forces.
An interactive segmentation method based on a genetic algorithm is de-
scribed by Cagnoni
et al.
[5]. In this method, the boundary contour of a region
of interest is manually drawn in one of the slices. The boundary contour is then
