Biomedical Engineering Reference
In-Depth Information
to share some of the stereotactic frame's independence of anatomy and
pathology. Maurer et al. demonstrated this independence by showing that
the differences between measurements of TRE on 100 patients and measure-
ments on a phantom were statistically insignificant.
15
The target feature was
a marker identical in design to the fiducial marker and implanted in the skull,
but it was not used in the registration process. The mean measured TRE was
found to be 1.0 mm for registrations of CT (3 mm slices) to physical space,
and 1.4 mm for registrations between MR (3 mm slices) and either of these
modalities.
To achieve this accuracy, a considerable effort was expended both in the
design of the markers and the design of algorithms to localize them. The
resulting FLE for this system is about 0.3 to 0.4 mm in CT images with 3 mm
slices, in MR images with 1 to 5 mm slices, and in physical space.
15-17
A simi-
lar study, also based on a skull-attached marker as target feature, has been
carried out for a similar skull marker system. The mean TRE for registrations
of CT (2 mm slices) to physical space was measured for 20 neurosurgical
patients and found to be 1.7 mm.
18
The theoretical dependence of TRE on posi-
tion suggests that attachment to the skull in each of these studies should pro-
vide an upper bound on RMS (TRE) (see Chapter 3), but the possibility of brain
motion, which is not reflected in these measures, limits the reliability of these
systems for validation to nonsurgical applications and skull base surgery.
19, 20
Target features are also appropriate for validation in nonrigid registration for
both intrapatient and interpatient applications. Miga et al., for example, used
an array of implanted target markers in a porcine model,
21
and Edwards et al.
22
used implanted electrodes
to validate their respective models of brain
deformation during surgery. Naturally occurring features may be used as
targets for validation as well. Woods et al., for example, employed cortical
landmarks comprising prominent gyri and sulci to validate a method for inter-
subject registration of PET and MR images.
23
Because these latter target features
are not point objects, their measure of error was the smallest distance between
voxels in the features. A similar method has been used in a cadaveric study to
validate rigid systems.
24
It is difficult to translate such measures into TRE.
6.3.1.3
Fiducial Marker Systems
A fiducial marker system can provide an excellent gold standard for both
intramodality and intermodality rigid registration. The transformations
some of these systems produce can provide submillimetric accuracy, and
actual patient images can be used. The primary disadvantage is that high
accuracy comes at the price of high invasiveness. Thus, the availability of
such images may be rare, and the set of available transformations is typically
limited.
The stereotactic frame provides such a system and has been used, for exam-
ple, to validate retrospective algorithms for PET-to-MR
25,26
and a point-based
system for neurosurgery based on skull-implanted markers.
18
Point-based
