Design and Implementation - Segmentation of Hand Bone for Bone Age Assessment - page 50

3. No complicated 'training' procedures: to prevent high dependency on availa-
bility of training samples of hand bone radiographs. However, simple parame-
ter tuning procedures without depending on availability of training hand bone
radiographs have to be established to capture the variations of uncertainties in
image nature.
4. Utilization of prior knowledge: to ensure the usage of available information
to optimize the result on hand bone segmentation. Besides, making use of
'by-products' of image pre-processing is preferable.
5. Relatively high resistance to noises: to improve performance of segmentation
despite the inevitable random signals in the hand bone radiographs.
6. Automated or minimum dependency on human interventions: to ensure objec-
tivity, to enable reproducibility and to avoid laboriousness.
7. Consistent accuracy: to ensure relatively high precision in segmentation on
resultant hand bone for subsequent processing in automated BAA system.
8. Resemblance to manual segmentation: to ensure a certain level of artificial
intelligence in the designed algorithm to emulate human visual perception.
9. No overdependence on certain image feature: to enhance segmentation robust-
ness under absence of any certain property such as intensity discontinuity or
edges.
10. Adaptability: to increase segmentation robustness under presence of variabil-
ity in different regions of hand radiographs.
11. Optimality: all parameters are chosen based on the direction of finding the
optimum solution and not arbitrarily pre-set. However, this criterion should
not violate the second criterion.
To facilitate the subsequent explanations on the proposed framework, hence-
forth, aforementioned desired property is referred as P1, P2, P3 … and so forth.
For example, the first property of contrast, illumination, orientation invariance is
referred as P1 and the tenth criterion of adaptability is referred as P10.
3.2 The Proposed Segmentation Framework
The proposed segmentation is a series of image processing procedures that is
specially tailored for solving the problem of hand bone segmentation. The proce-
dure begins with pre-processing to improve the input image properties so that it
is ready to go for the central processing algorithm namely the Adaptive Crossed
Reconstructed (ACR) algorithm. After that, it will undergo a feedback process to
assure the quality of the segmented image. The flowchart of the main procedures is
shown in Fig. 3.1 .
Pre-processing, main processing algorithm and quality assurance process repre-
sent the subroutines of the proposed framework. The details of each subroutine are
discussed follow.