Introduction - Segmentation of Hand Bone for Bone Age Assessment - page 7

sophisticated segmentation algorithms are scarce. In next sub-section, the dis-
cussion is on the background of the problem about the computerized system, the
approaches of other researchers that attempted to address the problem, and the
extent of the existing approaches in addressing the problem.
1.2 Background of the Problem
The current existing segmentation methods and frameworks either involve in
threshold settings or are too dependent on certain resources and image features.
This indicates that improvement on hand bone segmentation is necessary in order
to practically realize the fully automated computer-aided skeletal age scoring sys-
tem. Thus, this research is to explore this improvement aiming to establish an fully
automated segmentation framework that is accurate yet remains less dependent on
external resources.
1.3 Problem Statements
There are abundant of hand bone segmentation techniques found in the literatures,
but very few of which, if ever, is functioning as an effective and yet remains fully
automated. The research problem, therefore, is to explore the question: Is there
any method that can realize the goal of performing the automated segmentation
task that is relatively much more effective than fundamental segmentation tech-
niques but yet is unaffected by constraints such as training sets and human inter-
vention that are invariably pertain to sophisticated techniques?
The factors that associate with the problem are presented as follow:
1. The variability in hand radiographs deviates across different input sources and
different age groups of the subjects in radiographs. This variability impairs the
performance consistency or precision of segmentation technique or frameworks
once the input radiographs are not as expected.
2. Devoid of prior knowledge of computational algorithm in recognizing pat-
tern that can be easily perceived by human. As a consequence, most segmen-
tation framework necessitates explicit labors and hence this problem violates
automaticity.
3. The inherent bone intensity property in radiograph that stem primarily from the
variations in anatomical density of different parts of the hand bone. As a conse-
quence, two adverse properties for segmentation performance take place:
a. The overlapping range of pixel intensity for the cancellous bones, the soft-
tissue regions and the compact bones.
b. The non-uniformity within the same category of bone such as cancellous
bone or the cortical bone the radiograph intensity is not evenly distributed.