Biomedical Engineering Reference
In-Depth Information
Figure 8.
Sample individual selections and distributions
p
ind
u,c,i
(
x, y
)
for 2 control points
in Figure 1a. (a) Control point 1, user 1; (b) control point 1, user 2; (3) control point 2,
user 1; (4) control point 2, user 2. Each user made 10 selections of the control point for the
same image as in Figure 1a. The ellipses enclose 95% of samples along the principal axes
of
p
ind
u,c,i
(
x, y
)
. Reprinted with permission from the AAPM.
The output of the fourth and last step is the four distributions
p
VO
c,i
(
x, y
) that
comprise the virtual operator VO
i
for image
i
, as shown in Figure 9a for the image
of Figure 7a, with closeups for control points 1 and 2 shown in parts (b) and (c).
For each control point, the smaller ellipse indicates the spatial extent of
p
VO
c,i
(
x, y
),
whereas the larger ellipse encompassing it depicts
p
pop
c,i
(
x, y
). By design, both
ellipses have the same centers and axes. However, the extent of
p
VO
c,i
(
x, y
) is
smaller than the extent of
p
pop
c,i
(
x, y
) because
p
VO
c,i
(
x, y
) describes how a single
average user selects control point
c
, whereas
p
pop
c,i
(
x, y
) describes how all users
select the same control point. This shows that the population of users would have
more variability in their selections than would a single average user.
The ellipses representing all of the various distributions generally exhibit
lower variance in the direction perpendicular to the edge, and higher variance in
the direction tangential to the edge. This is reflected by the narrower width of
the ellipses in the direction perpendicular to the edge. In this direction, the sharp
transition in gray levels at the edge provides a clue to the user as to where to
select the control point; however, there is no such clue in the tangential direction,
resulting in increased variance. The variance in both directions is also higher for
control points 2 and 4 than it is for control points 1 and 3. As noted in Section 2.4.2
