Introduction - Segmentation of Hand Bone for Bone Age Assessment - page 10

models segmentation methods to develop an effective segmentation frameworks
consisting of several modules that are fully automated and independent from the
completeness of training samples and availability of skillful operators. As will be
shown, this proposed segmentation framework produces superior segmentations
yet it remains computationally feasible.
The main contributions presented in this topic are summarized as follow:
1. Extend the comprehensiveness of existing histogram equalization technique
by first assessing the current theoretical and technical architecture of existing
histogram equalization methods and then contribute the new insight to revo-
lutionize the conventional perception towards the ultimate goal of histogram
equalization by proposing the new histogram equalization framework; then,
based on the revolutionized insight, a holistic histogram equalization in terms
of luminance preservation, contrast, and detail preservation based on the Beta
function is developed to preprocess the hand bone radiograph serving the pur-
poses of standardizing and equalizing the non-standardized illumination among
radiographs that contain high variations in luminance, improving luminance
difference across edge borders in radiographs, reducing variations in lumi-
nance difference across edge borders among radiographs and most importantly,
enhancing the visual perceptual effect of ossification sites to improve the per-
formance in ossification localization and bone age assessment.
2. Extend the body of knowledge of anisotropic diffusion by exploiting the poten-
tial of being fully automatic and adaptive to input radiograph instead of being
subjectively tuned by operators to solve the problem of non-uniformity and
mitigate the undesired effect of overlapping intensity range. Both contributions
below have profound implication for advancing the field of anisotropic diffu-
sion and provide adequate ground for framework that requires autonomous ani-
sotropic diffusion.
a. Address the problem of manual diffusion strength by designing an auto-
mated diffusion strength scheme based on the diffusion coefficient function
of speckle reducing anisotropic diffusion (SRAD) grounding on the well-
founded statistical theory of the relation between sample variance and global
variance. The main strength is its computationally attractiveness and practi-
cal applicability.
b. Address the problem of manual scale selection by designing an automated
scale selection. The main strength of which compared to limited existing
automated scale selection schemes is that it requires no excessive filtered
image before making decision to halt the diffusion iteration.
3. Transform the manual and rigid adaptive division scheme into an automated
adaptive quadruple division scheme that embodies human cognitive abil-
ity. This transformation is significant not only in a narrow sense of hand bone
segmentation, but most importantly, it is of a generic breakthrough in the field
of image segmentation. This implicit modeling of human intuition and prior
knowledge solve the problem of high dependency on explicit human resources
in operating the algorithm. Furthermore, the scheme itself is a building block or