Result and Discussion - Segmentation of Hand Bone for Bone Age Assessment - page 126

As shown in Fig. 4.12 , the expected value of FOC is around 0.8 (0.7897).
This suggested strongly that averagely, most of the pixel intensity in actual bone
have been segmented, besides it shows as well, averagely, the false labeling in
ossification are in small numbers and hence it provide d evid ence of good seg-
mentation. Besides, it is noticed that the dispersion of FOC is within a narrow
range between 0.7130 and 0.8880. In terms of standard deviation, the data dis-
persion is only 0.0608 which sugge sted that the precisions of detected bone bor-
ders is very high. Similar to FOM , the segmentation framework could produce
consistent segmented hand bone regions with correct spatial location. In com-
par ison with radiographs before BRANEA algorithm, which has expected value
of FOC as 0.5179, the BRANEA algorithm has improved the segmentation for
52.48 % to 0.7897.
As shown in Fig. 4.13 , the expected value of F R AG is around 0.8 (0.8114)
which suggested that the number of absence or additional bones region residing
in ossification sites is not likely, otherwise the v alue wo uld have fluctuated below
0.5. Besides, it is noticed that the dispersion of F R AG is within a narrow range
bet ween 0. 750 and 0.8912. In terms of stan dard deviation, the expected deviation
of F R AG from expected value of F R AG amounts to only 0.0445 which sug-
gested that the precisions of detected bone borders is very high. This suggested
the consistency of the proposed segmentation framework in producing segmented
image with almost correct number of bones or less number of wrongly segmented
bones. In compari son wit h radiographs before BRANEA algorithm, which has
expected value of F R AG as 0.6997, the BRANEA algorithm has improved the
segmentation for 15.96 % t o 0.81 1 4.
It is observable that FOC , FOM and F R AG gained better segmenta-
tion results in radiographs of patients that have chronological l age below 9 years
old. From the figures above, it is obvious that FOC , FOM and F R AG gained
output less than their expected values. The main reason for this slight deteriora-
tion is due to the increment of structural complexity in hand bone radiographs
as the ag e of pa ti ents inc rease. However, overall, the small standard deviations
of FOC , FOM , F R AGa in radiographs across all ages discussed in previous
paragraphs suggested that the proposed segmentation framework possesses high
adaptability which in turn indicates that the proposed segmentation framework is
capable of adapting to different types of content in hand bone radiographs with
different number of ossification sites from categories of age 0 to 17.
Besides, it is observable that FOC , FOM and FRAG have strong degree of cor-
relation from the figures above. To quantify this correlation, Pearson's product-
moment coefficient, r XY as followings is used to compute the correlation:

n
i = 1

( Y i − Y )
( n − 1 ) S X S Y

X i − X

(4.14)

r ( X )( Y ) =