Biomedical Engineering Reference
In-Depth Information
parameters needed to be known beforehand to efficiently code them into a string.
Also if the range is large, the GA convergence can be slow.
1.5
Feature-Based Registration Techniques
Feature-based registration techniques rely on extracting and matching similar
features vector between two or more data sets in order to find corresponding
data points. So the two critical procedures involved in feature-based registra-
tion is
feature extraction
and
feature correspondence
. Feature representations,
which are invariant to scaling, rotation, and translation, are more desirable in
the matching process.
One of the most successful feature-based registration techniques, espe-
cially for multimodal registration, is by
maximization of mutual information
(MI) [25]. This technique works well for both MR and CT since they are infor-
mative of the same underlying anatomy and there will be mutual information
between the MR image and the CT image. Such a technique would attempt to find
the registration by maximizing the information that one volumetric image pro-
vides about the other. It requires no a priori model of the relationship between
the modalities, it only assumes that one volume provides the most information
about the other one when they are correctly registered.
Unfortunately, if initial transformation between the two modalities is un-
known, the MI will converge slowly. So, we will demonstrate how to use a com-
posite registration procedure that integrates another feature-based registration
technique, mainly a
surface-based
[26] registration technique, to estimate the
initial transformation. Surface-based registration techniques use features avail-
able on the data set surface mesh such as density or curvature. The surface-
based registration techniques work better for free-form surfaces, such as the
skin contours, while MI works better for voxel-based volumes. Such compos-
ite registration procedures have become recently the state-of-the-art in most
registration applications due to the fact that most feature-based techniques are
complementary in nature.
1.5.1
Surface-Based Registration Algorithm
In order for any surface-based registration algorithm to perform accurately and
efficiently, an appropriate representation scheme for the surface is needed.
