Information Technology Reference
In-Depth Information
Table 3 Relative Comparison
F
1
F
2
F
3
F
4
F
5
F
6
Klinder et al. [
1
]
No
>36.5
No
Yes
All
No
Mestmayer et al. [
4
]
Yes
>36
Yes
No
Speci
c
Yes
Proposed
Optional
<3
Yes
No
All
No
Table 4 Average execution time of the framework: The average time calculation is based on
12 VBs/96 CT slices
Framework stages
Execution time (seconds)
Spinal cord extraction
15.7
VB separation
18 (manual)/45 (automatic)
Initial segmentation
54.1
Shape registration
32.6
Final segmentation
46.8
Total
167.2 (when manual VB separation is considered)
The proposed framework is compared with two of very important spinal bone
related publications using features of each method. The features can be described as
follows: F1: User interactions, F2: Execution time (minutes) to run all steps respect
to segment 12VBs, F3: Extraction of spinal processes, F4: Vertebra identi
cation,
F5: Suitability to all or speci
c location of spinal column (such as thorocic, lumbar,
and etc.), F6: The BMD measurements. Since the direct comparison with these two
methods are very dif
cult, each feature is compared as shown in Table
3
. Although
the results are obtained using difference computer system for each method, the most
important contribution of this work is to segment VBs in very low execution times
with the acceptable segmentation accuracy. We maintain that the proposed method
can be applied in real time clinical studies.
It should be noted that VBs were manually separated in this test. The framework
take 167.2 s (less than 3 min) to segment 12 VBs
see Table
4
. The number of
slices affects the execution time. For the 2D/3D framework, the execution time is
related to the number of slices in the image. Some experimental images of 3D
results are shown on coronal and sagittal views in Figs.
26
,
27
,
28
and
29
.
—
2.5 Segmentation Using Euclidian Distance-based Shape
Model, LCG- based Intensity Model, and Asymmetric
Gibbs Potential-based Spatial Interaction Model
In this method, image modeling is a uni
ed approach, which is created by inte-
grating several of our previous and ongoing efforts in image modelling techniques.
The
first step is modeling shape variations using our new distance probabilistic
