Morphometric Data - An Invariant Approach to Statistical Analysis of Shapes - page 41

repeatedly, preferably four to five times, from a fixed specimen and the
average of the landmark coordinates be used for that specimen.
Determining the number of repeated observations per specimen or per
landmark depends primarily on two factors: 1) the distances between
landmarks (smaller distances needing more repetitions); and 2) the
landmark type (problematic landmarks requiring a larger number of
repetitions).
b) Measurement error for collecting coordinate data from CT
slice image
Use of computed tomography scans as a source for landmark coordi-
nate data requires an alternate error study. A study conducted in our
laboratory (Richtsmeier et al., 1995) evaluated the precision and
repeatability of locating anatomic landmarks in three dimensions
using CT slice images (each slice is 2D), and then validated these mea-
sures by comparing them to those taken using an already validated
measurement system (the 3Space digitizer). We do not need to be con-
cerned with issues of rotation and translation when comparing
measures taken from a single set of images because once a specimen or
patient undergoes a CT scan and the image is acquired, it is by defini-
tion fixed. However, we need to quantify any error that occurs in the
acquisition of the image, and this requires that we compare data col-
lected from various images of the same specimen.
In this study, 10 dry skulls underwent two episodes of CT scanning
resulting in two sets of CT images for each skull. Landmarks were col-
lected from each set of slice images during two separate data collection
episodes. This resulted in four landmark sets being collected for each
skull (two sets of landmark data from each image set). Differences in
data collected from the two CT image sets of a single skull are due to
error in digital recording by the CT scanner. Differences in data col-
lected at separate times from the same image are due to recording
error, which may have contributions from both the observer and the
data collection software. This design enabled separation of error due to
the imaging device from error in locating the landmarks on the images.
Finally, the same landmark data were collected directly from the same
ten skulls using the 3Space tabletop digitizer. Comparison of the land-
mark data sets collected from the CT scans and the digitizer
quantifieserror in the way that data are acquired and displayed by a
CT image and error in collecting data from the CT images (software).
The results of this study showed that three-dimensional landmark