Noor Syazana Bt Abd Ghani (20775) Senior Lect. Dr. Norashikin Bt Yahya
Student of Electrical & Electronic Engineering Electrical & Electronic Engineering Department
Universiti Teknologi Petronas Universiti Teknologi Petronas
31750, Tronog, Perak 31750, Tronoh, Perak
noorsyazanaabdghani@gmail.com norashikin_yahya@utp.edu.my
Figure 1 Block Diagram of Bone Age Analysis C) Feature Extraction and Feature
from Bone X-Ray Selection
By converting ROI images to the binary, we
A) Input Image split into two part which the epiphysis part and
The bone hand x-ray image dataset was diaphysis part. Rows scanning is used to measure
downloaded from the Kaggle.com website which the length of the epiphysis and diaphysis in order to
comprises the 200 testing images and 2000 training measure the minimum and maximum of column.
images in each folder. The folders consists of left Then, the ratio is measure by extracting that two
and right hand x-ray images. The dataset of images minimum and maximum of column.
X-ray are obtained with the CSV containing the
information on what is to be predicted (training D) Classification
folder) in such as age, gender, and some additional
information with reference to the standard of the The feature extraction will undergo the
Greulish and Pyle (GP). classification method. The proposed method
B) ROI Extraction basically using the SVM technique which perform
well when the dataset is noisier and that is why
The first stage to analyze the bone age SVM is commonly recommended for the
basically doing the preprocessing pipeline in order classification and nonlinear function approaches.
to extract the Region of Interest (ROI) from the x-
Another method is manually classified
according to the theory classification which we
manually create the table and classified the length
of feature extraction to the classes of BA.
IV. RESULT AND DISCUSSION
A) Input data Figure 2 Detection of Bounding Box
The input data is retrieved from the In Figure2 shows the detected bounding box
Kaggle.com website which contributed 12612 comes with value of confidence level. The
images of x-ray radiographs including the age data confidence level means that the detection of the
information that correspond to the classes of bone score returned as M- by- 1 vector of classification
age. The input data divided into two part in which scores. The larger score value will indicate the
to be use for the training and testing data. In order higher confidence in epiphysis detection and the
to get the ROI image. correct ROI which is the epiphysis of the middle
finger.
B) Preparation of Data
In our approach, we fixed the 100 training
image that consists of different classes of the bone
age to train into several deep networks architectures
to evaluate how the various bones contributed to the
model’s performance across the development of
bone stages. As the input data come from different Figure 3 ROI Detection on Testing Image
type in such the intensity of color, the size of the
radiograph and so on, there is compelling need for C) Extract the ROI Part
the reliable and general hand segmentation
For this stage, we removed unnecessary part
technique. However, mostly detection algorithm
by cropping out the exact detectable bounding box.
typically need in a large amount manually labeled
Use the command imcrop to crop the original
training set. Therefore, we use the image labeler
detectable box according to the rectangle of box
application in the Matlab to employ the technique
which indicates the size and position of the yield
called positive instances.
square shape (x, y, width, tallness) or can called as
In general, the positive instances/images is the coordinates. The trimmed picture incorporates
an iterative procedure where manual label that all pixels in the info picture that are totally or
combine with automated processing code. This will halfway encased by the square shape. The resulting
allows in quickly obtain accurate masks for the output image is 122-by-123.
testing image in the training set using Aggregated
Channel Feature Detector Method
As the ACF is the detection framework, this
framework uses to detect object at multi scale so in
this assessment, the x-ray image segmentation need
to train first by train ACF object detector to create
an ACF object or part that can detect the epiphysis Figure 3 ROI Detection
bone.
E)Enhancement of the ROI Epiphysis.
Getting the several technique of the image D) Measuring of Epiphysis and
enhancement approaches and comparing the Diaphysis
effectiveness as illustrated in Figure 4 and details
technique below: The ROI part is then divide into two part
which are the upper part and below part to get the
1. 1) Imadjust. This technique increase the contrast of ratio of length of development of the bones.
the image of the input intensity image. Thus, Basically, the binary image is to normalize
imadjust increase the contrast by default, saturates radiographs for a grayscale image. The image have
the bottom 1% and top 1% of all pixel value. the black background and white color in bones.
2. 2) Histreq. Histreq performs the histogram Image size varies considerably from a few numbers
equalization. Enhance the image of the contrast by to a few hundred pixels.
transforming value in intensity image output image
will approximately matched to the specific The epiphysis (upper part) of image, we
histogram which is the uniform distribution by calculated the pixel mean of the last 15 rows with
default. all columns image patches while the lower part of
3. 3) Adapthisteq. This technique perform the contrast ROI, we take the first 15 rows with all columns
limit adaptive histogram equalization. Adapthisteq image patches. By the row scanning, we extract the
is the other way instead of using the histreq change point which selecting the smallest and the
technique which operate on the small regions in the largest column.
image as well as tiles. It contrast of the each tiles to
get the output region approximately matches to the
specified histogram.
REFERENCES