Morphometric Data - An Invariant Approach to Statistical Analysis of Shapes - page 36

viewing a single slice, all features exist on a single plane, so all land-
marks within a slice will have the same Z coordinate. However, the
software may determine whether these features are placed on the
superior surface, the inferior surface, or projected midway between the
upper and lower faces of the slice. When a feature appears in two con-
tiguous slices, the Z coordinate of the corresponding landmark can be
estimated as halfway between contiguous slices. If the thickness of the
slice is greater than the dimensions of the pixel, the observer can
expect measurement error to be disproportionately greater along the
Z-axis as compared to the X- and Y-axes (see Richtsmeier et al., 1995).
Computerized reconstruction of CT images involves interpolation of
the pixels between slice images, thereby producing a three-
dimensional volume or surface of the form that has been scanned.
There are several reconstruction programs available (e.g., etdips
http://www . cc . nih . gov/cip/software/edtips/ ; Mullick and Venkatar-aman
et al., 1998); ANALYZE ( http://www . mayo . edu/bir/) ; (Robb and Hanson,
1995; Hanson and Robb et al., 1997; MEASURE (Barta and Dhingra et
al., 1997), VOXBLAST ( http://www . vaytek . com/VoxBlast . html ). Some
visualization and measurement software provide tools to determine
the location of landmarks in 3D space. A landmark can be located
simultaneously on the reconstruction, the axial images, and on recon-
structions orthogonal to the axial plane (e.g., sagittal, coronal). These
softwares and the hardware required to run them are becoming less
expensive as personal computer technology advances.
2.4 Reliability of landmark coordinate data
Studying the reliability of the data collection methods is essential
before any type of statistical analysis can be conducted. Moreover, a
specific instrument may require certain types of testing. Our general
advice is, know your instrument, talk with colleagues who have used
the instrument previously, and design a measurement error study that
will provide you with knowledge of the limitations of your data.
It is important to understand the various sources of error that can
contribute to what is measured. Kohn and Cheverud (1992) define pre-
cision as the average absolute difference between repeated measures
of the same specimen. Precision of a method of data collection is mea-
sured as the variability among repeated measures of the same
specimen. Lack of precision results in variability among repeated mea-
sures of the same specimen and has two components: 1) observer error