Conclusion and Future Works - Segmentation of Hand Bone for Bone Age Assessment - page 130

and uneven texture within soft-tissue region, within the cancellous bone, within
the cortical bone and within the radiographic background. This smoothing process
is conductive in mitigating the inferior segmentation effect resulting from non-uni-
formity within specific region and overlapping distribution of intensity range.
The quadruple division has been proposed and inserted into the segmentation
framework to instill a certain extend of human cognitive ability by using the rule-
based fuzzy inference system to achieve the forth objective. This scheme suc-
cessfully searches optimized size in applying the central fundamental algorithm
automatically according to the texture information of segmented region without
presetting the size and number of division. This scheme realizes the requirement
of the framework in terms of automaticity and adaptability.
A post-processing quality assurance scheme has been proposed and applied
in the segmentation framework to achieve the fifth objectives. As the result has
shown, this scheme is capable of supplying the ACR algorithm to the preprocessed
hand bone radiograph with the most suitable size of block by modeling human
knowledge implicitly. The process is vital in assessing the segmented image and
determines whether there is a need to restore the lost detail, eliminate unwanted
segmented regions or remain as it is.
Generally, the contribution of this study is that it adds substantially to the
understanding of the possibility to solve a complicated segmentation problem
by incorporating merely any relatively simple algorithm together with a series of
customized modules containing a certain level of prior background knowledge of
human. Lastly, the main contribution of the proposed technique is not merely to
establish a novel segmentation framework to specifically resolve the hand bone
segmentation problem but to introduce a new generic segmentation framework
with the mentioned concept that has high generality to be applied and modified in
different biomedical image processing applications. The segmentation framework,
as a whole, fosters the insight about recognizing the inherent limitations of tech-
niques and input information and then knowing how to extract their relational pat-
terns via critical analysis to derive relevant measures for the purpose of rendering
the optimized 'environment' and 'material' for a relatively fundamental algorithm
to leverage the performance of which to a level that is comparative to sophisticated
algorithms that involve unpractical constraints.
5.2 Future Works
This research has opened up new research questions in need of further investiga-
tion. Firstly, the future exploration on optimum kernel size in filtering processing
or edge detection process is necessary to optimize the adaptability and automatic-
ity on various dataset from different input sources. The kernels play a pivotal role
analogous to human 'eyes' to process the local information in image. It remains a
challenge to choose the optimum size of kernel that able to emulate human cog-
nitive visual ability. Secondly, it would be interesting if the priority of multiple