Biomedical Engineering Reference
In-Depth Information
considered an initial contour in the subsequent slice and the contour is refined
by a genetic algorithm using image information. The refined boundary is then
considered an initial contour in the next slice and the process is repeated until all
slices in a volumetric image are segmented. Interactive segmentation methods
provide varying levels of user control. The control can be as little as selecting
a contour among many [19] or as much as manually drawing a complete region
boundary. Methods that require a lot of user interaction are highly reliable, but
they also have a high interuser variability. On the other hand, methods that re-
quire very little user interaction are not as reliable, but they have a low interuser
variability. A survey of interactive segmentation methods providing different
levels of user control is given by Olabarriaga and Smeulders [23].
The new idea introduced in this paper is to use the capabilities of a computer-
aided design system to quickly and effectively refine the result of a 3D segmen-
tation, just like editing a 3D geometric model. By having a mental picture of an
object of interest and viewing the information present in a volumetric image, the
user interactively modifies the result of an automatically obtained segmentation
until the desired shape is sculpted. This is achieved by representing the region
by a parametric surface and overlaying the surface with the volumetric image.
Then, the user views both the image and the surface together and modifies the
surface until the satisfactory region is obtained.
We assume an automatic segmentation method that correctly finds most
parts of a region of interest is available. The capability introduced in this paper
enables the user to revise parts of the region that are believed to be inaccurate.
This revision is achieved through a mechanism that sculpts a desired shape from
a rough initial one. The proposed method is not the same as a dynamic snake
model that creates a desired shape by interactively changing the external forces
that guide the snake [20]. Rather, it is based on a parametric surface fitting and
editing model.
7.2
The Computer-Aided Design System
We assume a volumetric image has been segmented and a region of interest
has been extracted. We also assume the given region is composed of connected
voxels that represent the bounding surface of an object of interest. We will call
such a region a
digital volumetric shape
,ora
digital shape
. In the following, a
