Biomedical Engineering Reference
In-Depth Information
and vascular malformation may lead to brain stroke, which is the third leading
cause of death and the main cause of disability. An accurate model of the vascular
system from MRA data volume is needed to detect these diseases at early stages
and hence may prevent invasive treatments. A variety of methods have been
developed for segmenting vessels within MRA. One class of methods is based
on a statistical model, which classifies voxels within the image volume into
either vascular or nonvascular class for time-of-flight MRA [56]. Another class of
segmentation is based on intensity threshold where points are classified as either
greater or less than a given intensity. This is the basis of the isointensity surface
reconstruction method [57-59]. This method suffers from errors due to image
inhomogeneities in addition; the choice of the threshold level is subjective. An
alternative to segmentation is axis detection known as skeletonization process,
where the central line of the tree vessels is extracted based on the tubular shape
of vessels [60]. Other approaches for MRA vessel segmentation are the manually
defined seed locations for segmentation [61].
In this section, we use level set method for image segmentation to improve
the accuracy of the vascular segmentation. This work is a supervised classifi-
cation which means that the number of classes and the class distribution are
assumed to be known. Usually, the class distribution is assumed to be Gaussian
with known mean and variance. In [53], classes were assumed to be phases sep-
arated by interface boundaries where each class has its corresponding level set
function. A set of functionals were developed with properties of regularity. The
level set function representation depends on these functionals. Each class oc-
cupies certain areas (regions) in the image. The level set function is represented
based on the regions i.e. it is positive inside the region, negative outside, and
zero on the boundary. The classes have no common areas i.e., the intersection
between classes is not allowed. The sum of lengths of the interfaces between
the areas is taken in consideration. The functionals are dependent mainly on
these properties and they are expected to have a local minimum which is the
segmented image. The change of each level set is guided by two forces, the min-
imal length of interfaces which is the internal force and the homogeneous class
distribution which is the external one.
A PDE guides the motion of each level set. This work saves the manual
initialization of level set functions [62]. Bad initialization for these functions
makes the segmentation fail. Automatic seed initialization is made for each
slice of the volume by dividing the image into windows, and based on the gray
