Morphometric Data - An Invariant Approach to Statistical Analysis of Shapes - page 42

data collected from CT slice images are internally consistent and pre-
cise. It was noted, however, that error along the Z-axis (the axis along
which the table is moving during image acquisition and which defines
the slice thickness) is greater than error along the other two major
axes. A subsequent study (Valeri et al., 1998) found error along the Z-
axis to be reduced when the same data were collected from
three-dimensional reconstructions of computed tomography slice
images.
When collecting landmark data from 3D CT reconstructions, tradi-
tional landmarks that do not depend upon the orientation of the
specimen are desirable because they exclude a potential source of error
(Richtsmeier et al., 1995). Adding the coordinate system to the defini-
tion of a landmark adds another level of potential error to landmark
identification. An example of a landmark that is dependent upon a spe-
cific orientation is glabella, defined as the most anterior projecting
point on the frontal bone. The position of this landmark relative to
other features of the skull will shift depending upon the orientation
(coordinate system) of the specimen, and the orientation needs to be
defined in order to collect comparable data from other specimens.
We use the term “local coordinate system” to refer to the coordinate
system inherent to the object under study. When collecting landmark
data from an object sitting on a digitizer, the local coordinate system is
defined on the basis of the position of the object. If the object is moved,
the local coordinate system changes. This is why an object must remain
stable during data collection using a digitizer. However, when slice
images or three-dimensional reconstructions of images (MRI or CT for
example) are used for data collection, the local coordinate system is
captured along with the image and remains inherent to the image
regardless of any movement of the reconstructed image within most
software programs. In effect, the exact coordinates for any given land-
mark (e.g., lambda) are the same whether collected from the 3D
reconstruction oriented in the Frankfort horizontal position or from
that same reconstruction turned upside-down and tilted at an oblique
angle. The imaged specimen can be moved and tumbled through space
during data collection.
We emphasize that these are only examples of measurement error
studies for landmark data collected using specific devices. With each
new instrument, researchers should conduct specialized studies of
measurement error in order to assess the data collector's competency
and the instrument error.