Biomedical Engineering Reference
In-Depth Information
[44] of the CP and AP contours, and the OPT contour on the left was obtained
from the CP contour alone. These examples show qualitatively that our algorithm,
which combines the CP and AP contours, generates a result that is closer to manual
segmentation than any of the other slice-based 3D segmentation approaches.
Table 3.
Prostate Volume Measurements (cm
3
) Associated
with Study 1 for the Manual Segmentation
Method (MAN) and the Four Slice-
Based Segmentation Methods
Reject Null
Hypothesis
H
o
:
σ
i
=
σ
1
σ
i
σ
1
σ
i
Method
V
i
1
V
i
2
V
i
3
V
i
4
V
i
5
V
i
F
=
p-value
1
20.9 30.9 21.9 24.2 50.7 29.7 1.21
-
-
-
2
16.0 24.8 16.6 18.4 41.8 23.5 1.92
1.59
0.1550
No
Volume
(cm
3
)
3
16.7 27.1 17.5 20.1 42.8 24.6 2.52
2.08
0.0545
No
4
15.4 22.9 13.0 18.6 39.0 21.8 5.19
4.29
0.0010
Yes
5
19.3 29.0 18.5 21.7 46.6 27.0 1.21
1.00
0.5000
No
SB = slice-based segmentation method described by Wang et al. [41];
CP = clockwise propagating; AP = anti-clockwise propagating;
OPT = optimal AR model-based segmentation, which are labeled respectively from methods 1 to 5.
The prostates depicted in Figure 14a-f are labeled prostates 1 to 5, respectively. For each segmentat
ion
method
i
and prostate
j
, segmentation was performed five times and the volume computed.
V
ij
was computed by taking the mean for the five
re
peated volume measurements. The average volu
me
associated with each segmentation method
i
,
V
i
, was computed by averaging the mean volume
V
ij
associated with prostates 1 to 5. Using the one-way ANOVA model, the variance of the volume
measurement for each segmentation method
i
,
σ
i
, was computed and tabulated. The
F
-test was
used to compare the estimated variance associated with each algorithm and manual segmentation.
The
p
-value, the probability that the two population variances are the same while the ratio of two
estimated variances are greater than the computed
f
-value associated with each comparison set, was
also tabulated.
Table 3 lists the mean volume measurements of the prostates, with the MAN,
SB, CP, AP, and OPT segmentation methods label
ed
i
= 1 to 5 respectively. For
each segmentation method
i
, the average volume
V
i
and its associated variance,
σ
i
, wwere calculated from the multiple segmentations. By computing the ratio
between the variances associated with the algorithm segmentation methods,
σ
i
,
and the variance associated with the manual segmentation method,
σ
1
, and using
the
F
-test, we found, with a 95% confidence interval, that the variance of the AP
method,
σ
4
, is higher than that with the MAN method, whereas the variances as-
sociated with the other three algorithm segmentation methods are not statistically
different from the manual method. Using a one-way ANOVA, we have also cal-
culated the
p
-value: the probability that the two population variances are the same
