Biomedical Engineering Reference
In-Depth Information
Mean of fuzzy quadruple scheme
Expected value of fuzzy quadruple scheme
Mean of rigid scheme with row=4, column=4
Expected value of rigid scheme with row=4, column=4
Mean of rigid scheme with row=2, column=2
Expected value of rigid scheme with row=2, column=2
Mean of rigid scheme with row=3, column=3
Expected value of rigid scheme with row=3, column=3
Mean of rigid scheme with row=1, column=1
Expected value of rigid scheme with row=1, column=1
1.2
1
0.8
0.6
0.4
0.2
0
0
2
4
6
8
10
12
14
16
18
Age Group
Fig. 4.7 Comparisons of segmentation accuracy between manual rigid schemes with various
divisions and proposed automated fuzzy quadruple division scheme using the mean of FOM and
expected value over all age group
was as expected. The proposed division scheme had optimized the adaptive block
image size for ACR segmentation algorithm and hence the superior results were
obtained. This result was intuitive according to the qualitative observation of seg-
mented bone shown in Fig. 4.6 that most of the critical edges of bones structures
were retained in Fig. 4.6 f in comparison to the Fig. 4.6 b-e where most of hand
bone outlines were inferiorly segmented, most importantly, the edges of ossifica-
tion sites that carried higher weights were missing.
Figure 4.8 illustrated the segmentation accuracy comparison between rigid
divisio n schemes and the proposed quadruple division scheme. The metric
FOC quantified the degree of discrepancies in labeling of gray level between
reference radiograph's pixels and the segmented radiographs over all age group
among rigid division schemes and the proposed quadruple division scheme. The
above experimental results have showed that the proposed division scheme pos-
sessed apparent superior performance of pixel labeling in comparison to other
rigid division scheme. Specifically, the expected value of proposed division
scheme over all age groups was 0.5179, followed by 0.2755 from rigid scheme
of row = 4, column = 4, 0.2156 from rigid scheme of row = 2, column = 2,
0.1454 from rigid scheme of row = 3, column = 3, and 0.1906 from rigid scheme
of row = 1, column = 1. The overall average of these rigid schemes in terms of
FOC was 0.2068. In terms of percentage, averagely, the proposed scheme per-
formed 150.4 % better than rigid schemes. Such an improvement was expected
from the result of Fig. 4.7 . This improvement indicated that the false labeling of
proposed scheme in performing the ACR algorithm in the problem of hand bone
 
Search WWH ::




Custom Search