Biomedical Engineering Reference
In-Depth Information
data at different scales are the most robust. These methods start by blurring
the data volumes, both in the given brain and the target data set, to remove
detail before estimating an initial registration. This first result is used as a
starting point for a second registration procedure using slightly less blurred
data to refine the initial fit. When this procedure is repeated a number of times,
reducing the data blur and refining the spatial mapping function, the result is
both reliable and robust. This strategy has the added benefit of reducing the
computational time required to compute a registration at high resolution,
since the initial stages can be computed quickly on highly blurred data and
the initial fits target the result and reduce the effective search space.
14.2.5
Other Brain-Based Reference Spaces
It is important to note that while versions of the Talairach coordinate system
have become the
standard in much of the brain mapping field, other
reference spaces exist that are applied for specific types of analysis.
In some studies, a single brain is selected as the target for registration and
standardization. This method suffers from the problems associated with
using a single brain (mentioned in Section 14.2.1.1) without the advantages
of an atlas-based coordinate space (see Roland and Zilles for an overview of
the uses of brain atlases as a research tool
de facto
63
). In neurosurgical planning of
stereotactic interventions, the reference space most often used is that of the
Schaltenbrand and Wahren atlas,
4
used to derive the coordinates of deep
brain targets.
In contrast to 3D registration, 2D methods concentrate on registering
pairs of surfaces by finding a mapping between the surfaces. Structures on
the surface are used to constrain the mapping. The surface most commonly
used is the outer cortical surface, i.e., the gray-matter
CSF interface. How-
ever, one could equally well use the white-matter
gray-matter boundary or
a medial surface inside the gray matter. Sandor,
64
65
Thompson,
and
66
Davatzikos
each developed methods that begin by automatically extract-
ing the cortical surface followed by manual identification of a small number
of sulci that are used to constrain the registration. While these techniques
usually maintain the 3D geometry of the 2D surface, the methods devel-
oped by Van Essen and Drury at Washington University can be used to
extract and
the cortex onto a 2D surface where one can apply surface-
based coordinate systems to analyze structure and function.
flatten
15
The main dif-
ficulty with this type of technique is the spatial distortion incurred by flat-
tening a 3D surface. The Washington University group has minimized
distance and area errors by inserting
into the surface. They have
applied the technique mostly to the visual system. The group at Harvard
has also developed a 2D coordinate system to identify regions of interest
based on automated surface extraction and flattening methods.
cuts
67,68
The
extracted surface has a sphere topology, so a spherical coordinate system
can be used. This method has been used to generate an average cortical
