imperfections, the error in the artificial point represents
fittihg error, and the error in the anatomic point
represents fitting and pointing error.
This process resulted in the boundaries being fitted
together with four, five, and six parameter transformations,
using all boundary pixels and a subset of pixels. The
results are shown in Table 1.
Residuals from Feature Transformations (Pixels at Generated Point)
Transformation Film Transformed Mean
No. of Boundary
Parameters | Pixels 2 —»1 3 -—»1 4 —> 1
20 0.79 + 0.42 | 0.96 + 0.37 | 1.01 + 0.80 0.92 + 0.12
4 10 0.77 + 0.37 | 0.93 + 0.41 | 0.98 + 0.48 | 0.89 + 0.11
6 0.82 + 0.37 | 1.03 + 0,51 | 1.07 + 0.51 | 0.97 + 0.13
20 0.69 + 0.32 | 0.86 + 0.17 | 0.51 + 0.31 0.69 + 0.24
5 10 0.74 + 0.44 | 1.00 + 0.26 | 3.31 + 2.82 | 1.68 + 1.42
6 1.19 + 0.56 | 2.64 + 2.65 | 2.96 + 1.65 | 2.26 + 0.94
20 0.55 + 0.29 | 0.74 + 0.15 | 0.25 + 0.08 | 0.52 + 0.25
6 10 0.80 + 0.56 | 0.97 + 0.42 | 0.73 + 0.16 0.83 + 0.12
6 1.03 + 0.82 | 1.31 + 0.87 | 1.21 + 0.04 1.18 + 0,14
Mean 0.67 + 0.27 | 1.16 + 0.58 | 1.34 + 1.06 | 1.05 + 0.35
Table 1 - Anatomic Feature Boundary Transformations
4.2 Analysis
A number of conclusions can be drawn from the boundary
matching tests, All statements of significance of
difference made in this analysis were confirmed by t tests
at the .01 significance level, unless Stated otherwise. The
data in Table 1 are the mean results obtained from all four
anatomic points, averaged by type of transformation (reading
down the table) and by film (reading across the table).
The best overall results were obtained With six parameter
matches on all of the boundary points. However, the
difference between the five and six parameter solutions is
only significant at the .05 level. This indicates that the
ability to scale independently in x and Y is a more
important factor than non-orthogonality of the axes, when a
single image is rotated and fitted,
- 178: =~