Biomedical Engineering Reference
In-Depth Information
great care. Simulation of intrapatient motion should be based on tissue
mechanics or growth models, and simulation of interpatient variation must
be anatomically sound. Models based on splines and other convenient math-
ematical formulations will tend to favor registration systems that employ
transformations with similar formulations, and therefore are highly suspect
as standards for nonrigid validation. The overriding problem is that, unless
constraints of some kind are placed on the motion, the number of parameters
necessary to specify the physical transformation of a continuous, nonrigid
object, or between two such objects, is large. In fact, absent such constraints, the
number is infinite. When the constraints of rigidity are inappropriate, then the
choice of constraints becomes the overriding issue in the registration.
6.3.1.2
Target Features
Rigorous attempts to use experimental statistics to predict clinical registra-
tion accuracy were first applied to the stereotactic frame.
14
The frame, which
is rigidly attached to bone, usually the skull, is used to effect a registration
between the physical patient and an image of that patient, typically for guiding
biopsy needles or radiation beams to lesions within the head. In these applica-
tions the head is held motionless in the scanner and (optimally) in the same ori-
entation relative to gravity as during treatment. Furthermore, major resection
is not involved. Thus the relationship between skull and brain remains rigid.
The frame is in fact appropriate only for rigid anatomy because all positioning
is measured relative to the frame, a (removable) part of which is visible in the
image. The rigid combination of frame and anatomy makes the registration
problem much easier than the general one and potentially more accurate,
because registration cues can be taken from a device that is built expressly to
provide such cues.
Use of the frame for registration also makes the job of validating accuracy
easier and potentially more reliable. Because neither the anatomy nor the
pathology are involved in frame-based registration, its accuracy is largely
unaffected by them and can be determined by means of carefully controlled
experiments on phantoms. In 1992, Maciunas et al. carried out a large scale
experiment of this kind on several frames.
14
They employed a phantom to
which, in addition to the frame, imageable markers were attached. These
markers served as
meaning that they played the role of the tar-
gets of therapy. By localizing a given target both in image space and physical
space, applying the transformation determined by the frame to the localized
point, and then measuring the disparity between the registered point and the
physical point, TRE can be estimated. It can thus be seen that the target fea-
ture is in effect a gold standard that provides a transformation at only one
point. Using this gold standard, the study found that a mean TRE of 2 mm
was possible.
Maurer et al. in 1997
target features,
15
also used the target-feature approach to estimating
TRE for a registration system based on specially designed, skull-implanted
fiducial markers. The accuracy of any marker-based system can be expected
