Result and Discussion - Segmentation of Hand Bone for Bone Age Assessment - page 118

4.3.2.1 Evaluation on Automated Fuzzy Quadruple Division Scheme
The defined metrics have been implemented to evaluate the outcome of quadru-
ple division scheme to justify its utility in the segmentation framework. Firstly,
the qualitative assessment is conducted to visually observe the effect of the
scheme by inspecting the segmented bone radiographs. Comparisons are made
between the rigid adaptive division schemes with the proposed fuzzy quadruple
division scheme on a quadrisected radiographs to inspect the differences. The
input images are processed by the similar pre-processing in the proposed seg-
mentation framework (constant variables), then they were implemented sepa-
rately manually set rigid adaptive division scheme with different division layers
and fuzzy quadruple division scheme (manipulated variables), the visual effects
of those images are inspected and analyzed on qualitative analysis using human
visual system and metrics are implemented to conduct the quantitative analysis
on a large image sets to further justify the qualitative analysis (response vari-
ables). The visual segmentation result of TR-2.1 was shown in Fig. 4.6 to illus-
trate the differences.
Figure 4.6 a denotes the original quadrisected radiograph; Fig. 4.6 b denotes the
TR-2.1 that was implemented with no division at all; Fig. 4.6 c denotes the TR-2.1
that was implemented with division of 2 rows and 2 columns; Fig. 4.6 d denotes
the TR-2.1 that was implemented with division of 3 rows and 3 columns; Fig. 4.6 e
denotes the TR-2.1 that was implemented with division of 4 rows and 4 columns;
Fig. 4.6 f denotes the TR-2.1 that was implemented with the proposed automated
fuzzy quadruple division scheme.
The effects illustrated in Fig. 4.6 were interpreted in a few perspectives.
From the perspective of delineated bone structures, Fig. 4.6 b-e failed to deline-
ate the bone structures, meaningful outlines of hand bones could hardly been
observed, however, the Fig. 4.6 f that had undergone quadruple division scheme
produced segmented hand bone that could at least provide noticeable hand bone
shape. From the perspectives of segmented bone structures region, Fig. 4.6 b-e
showed that rigid division scheme produced excessive noises and failed to
label correctly for most of the pertinent regions of ossification sites, whereas,
Fig. 4.6 f showed that via the proposed division scheme, most of the regions of
desired bones structures were obtained. The reason lies in the adaptability of
both scheme: the rigid scheme divides the radiograph regardless of the informa-
tion that was contained inside the input and also regardless of the previous ACR
algorithm; on the contrary, the superior performance of fuzzy quadruple scheme
is due to the consideration on the information of the inputs and also the infor-
mation of the output of previously implemented ACR algorithm to dictate the
number and size of divisions. Note that all the images had not been processed
by the quality assurance step; in next sub-section, the refinement of segmented
hand bone shown in Fig. 4.6 f is discussed using proposed quality assurance
step.
Figure 4.6 showed only an arbitrary example of the processed radiograph to
convey the qualitative visual effect of the proposed scheme compared to rigid