Information Technology Reference
In-Depth Information
solution for a given initial segmentation and shape model. Now, the objective is to
optimize the following equation to maximize the likelihood energy functional.
Algorithm 2 shows the proposed segmentation process using a new ICM method.
Algorithm 2: Optimization of Three Models
While
do
For all
do
Update
by the value of
which maximizes
End for
Increase
End while
Note: It should be noted that X
¼
Sx
þ
Tr is any transformed pixel,and
X
is the
pixel domain in the image.
L
ð
I
;
d
;
f
;
T
Þ
¼
logpðfÞ
ð
I
j
f
Þþ
logpðfÞ
ð
f
Þþ
logpðfÞ
ð
d
j
f
;
T
Þ:
ð
25
Þ
2.4.7 Experimental Results-CT Data Population
The training and testing images were acquired from GE LightSpeed VCT, Toshiba
Aquilion, and Imatron C-150 CT scanners with an in-plane resolution range of
0.63
0.98 mm and a slice thickness of 0.63
3.00 mm. For the testing stage, 18
-
-
patient data sets, of which 10 are from female (
'
F
'
) and 8 are from male (
'
M
'
), and a
phantom are examined in this study. There are 16
512
voxels. The proposed algorithm is tested on 932 CT slices/66 VBs which are
obtained from different spine bone regions (i.e. lumbar, thoracic, and etc.). In the
datasets, the number of visible VBs changes from 2 to 8. The data sets are also
categorized as
-
96 axial slices with 512
'
healthy
'
(H) and
'
with low bone mass
'
(L) with respect to their
calcium absorbtion. The experiments are run on 7
data sets. The ages
of the test subjects varies between 38 and 76 years with an average age of 61.3
years with 12.2 years standard deviation.
'
H
'
and 11
'
L
'
2.4.8 Experimental Results-Segmentation Measurements
Figure
17
shows the region of true positive (TP), true negative (TN), false positive
(FP), and false negative (FN). In this
figure, the reference region represents the
ground truth which is veri
ed by a radiologist. The test region represents the
automated segmented region. For the ESP, the segmentation quality is measured
