Biomedical Engineering Reference
In-Depth Information
Each 3D image contained 294
×
184
×
176 voxels of approximate size 0.34
×
×
0.34
0.31 mm.
4.1.2. Evaluation of slice-based segmentation
In evaluation of the accuracy of the slice-based segmentation method, the
manual volume estimates generated by the parallel re-slicing method,
V
mp
, and by
the rotational re-slicing method,
V
mr
, were used as the “true” volume estimates.
Thus, the percentage error
E
k
in the volume estimate
V
k
is given by
V
k
V
mk
−
1
E
k
=
×
100%
,
k
=
p, r.
(9)
While both methods can produce useful prostate volume estimates, it is in-
structive to examine which is better for prostate segmentation. This can be ex-
amined by investigating the accuracy and precision of the parallel segmentation
method relative to manual planimetry by calculating the mean and standard devi-
ation of the parallel error
E
p
, given by Eq. (10). Similarly, the mean and standard
deviations for the rotational error
E
r
can also be calculated. The systematic “er-
ror”
E
rp
resulting with the use of the rotational segmentation method rather than
the parallel segmentation method is then given by
V
mr
V
mp
−
1
E
rp
=
×
100%
.
(10)
Table 2.
Summary of Prostate Volume Errors and Editing Rate for Parallel
and Rotational Segmentation Algorithms (mean
±
standard deviation).
Segmentation Method Volume error
Volume
|
error
|
Editing rate
Parallel
-5.42
±
4.43% 6.49
±
2.07%
19.7
±
5.4%
Rotational
-1.74
±
3.51% 3.13
±
2.06%
14.4
±
5.4%
Errors in the volume estimates are measured relative to manual planimetry, done
in the same geometry as the corresponding segmentation method. Reprinted with
permission from the AAPM.
4.1.3. Evaluation results of slice-based segmentation
A comparison of the volume error and editing rate for the parallel and rota-
tionally re-slicing segmentation methods are listed in Table 2. This table shows
