Biomedical Engineering Reference
In-Depth Information
12.2.1.1
Error in Registration of Images to Physical Space
The clinical accuracy requirement of image guidance is one of the most
important criteria when deciding which technology to apply to a particular
application. Engineers use the word “error” to define the limits to accuracy
of a system. Surgeons prefer “accuracy” because of the wider connotation of
the word “error” in surgery. Some insist that their system never introduces
any error. In this chapter we stick with the engineers' definition.
Different definitions of registration error have caused some confusion in the
past, so before introducing the different technologies it is worth defining the
relevant measures of error. The terms used here were introduced in Section
3.4.1.2 of Chapter 3, and discussed in detail in Section 6.2.1 of Chapter 6.
Image-guided surgery relies on identifying corresponding features both in
preoperative images and the physical space of the patient in the operating
room or interventional suite. An error will be associated with locating these
features in the images. There will also be an error in identifying the physical
location of these points in the patient. If these features are defined as points in
both, such as the centers of spheres or cylinders
2,3
4
or intersection of two lines,
the error in correspondence is often referred to as the fiducial localization error
The corresponding features are used to compute a coordinate transfor-
mation between the images and physical space. This coordinate transforma-
tion is then used to predict where a point in physical space—the “target,”
indicated, for example, with a hand-held pointer—is located in the preopera-
tive images.
The error in predicting location of the target in the preoperative image is
often referred to as target registration error (TRE). Fitzpatrick et al. have
derived a formula for TRE from the point distribution in fiducial point-based
registration assuming that the FLE is isotropic and randomly distributed.
.
5
(
FLE
)
5
Point-based registration using the orthogonal Procrustes method also com-
putes a residual root-mean-square (RMS) error on aligning the points, which
is sometimes termed the fiducial registration error (FRE). It is important to
remember that FRE does not estimate TRE. In particular, discarding points
until the FRE drops below some threshold is a very poor way of ensuring a
good TRE.
A number of sources of error contribute to the overall system error of any
surgical navigation system. These include:
• Error in locating features for registration in the images. This is
determined by the size and shape of the features and the spatial
and contrast resolution of the images.
• Scaling errors and other geometric distortions in the images.
• Error in locating the same features in the physical space of the patient.
•
Any relative movement of these features between imaging and
intervention. For example, skin is mobile, so skin markers can
move.
