Classification of Long Bone X-ray Images using New features and Support Vector Machine

Abstract

Bones are protecting many organs in the human body such as the lungs, brain, heart and other internal organs. Bone fracture is a common problem in human beings and may occur due to the high pressure that is applied to the bones as a result of an accident or any other reasons. X-ray (radiograph) is the noninvasive medical experiment that helps doctors diagnose and present medical conditions. X-rays represent the oldest and most often used kind of medical imagery. Medical image processing attempts to enhance the bone fracture diagnosis processes by creating an automated system that can go through a large database of the X-ray images and identify the required diagnosis faster and with high accuracy than the regular diagnosis processes. In this paper, the lower leg bone (Tibia) fracture is studied and many novel features are extracted using various image processing techniques. The purpose of this research is to use new investigated features and classify the X-ray bone images as a fractured and non-fractured bone and make the system more applicable and closer to the user using the Graphical User Interface (GUI). The Tibia bone fracture detection system was developed in three main steps: the preprocessing step, feature extraction using wavelet analysis, gradient analysis, principal components (PCA), and edge detection methods and classification using Support Vector Machine (SVM). The results were produced using four possible outcomes from the confusion matrix which are TP, TN, FP, and FN. The classification process was repeated using different feature groups at a time and the resultant accuracies were ranged between 70%-80%.

Full text

Amani Al-Ghraibah et al., International Journal of Advanced Trends in Computer Science and Engineering, 10(3), May – June 2021,
1494 – 1500
1494
ABSTRACT
Bones are protecting many organs in the human body such
as the lungs, brain, heart and other internal organs. Bone
fracture is a common problem in human beings and may
occur due to the high pressure that is applied to the bones as
a result of an accident or any other reasons. X-ray
(radiograph) is the noninvasive medical experiment that
helps doctors diagnose and present medical conditions. X-
rays represent the oldest and most often used kind of
medical imagery. Medical image processing attempts to
enhance the bone fracture diagnosis processes by creating an
automated system that can go through a large database of the
X-ray images and identify the required diagnosis faster and
with high accuracy than the regular diagnosis processes. In
this paper, the lower leg bone (Tibia) fracture is studied and
many novel features are extracted using various image
processing techniques. The purpose of this research is to use
new investigated features and classify the X-ray bone
images as a fractured and non-fractured bone and make the
system more applicable and closer to the user using the
Graphical User Interface (GUI). The Tibia bone fracture
detection system was developed in three main steps: the
preprocessing step, feature extraction using wavelet analysis,
gradient analysis, principal components (PCA), and edge
detection methods and classification using Support Vector
Machine (SVM). The results were produced using four
possible outcomes from the confusion matrix which are TP,
TN, FP, and FN. The classification process was repeated
using different feature groups at a time and the resultant
accuracies were ranged between 70%-80%.
Key words: Bone fracture detection, Classification of bone
fracture, Features extraction, Support Vector Machine
(SVM), Tibia bone X-ray images.
1. INTRODUCTION
Tibia fractures are the most common long bone fracture
accounting for more than 20% occupancy of hospital wards.
On average 26 tibia fractures occur per 100,000 populations
per year. Several accidents require health care experts
to analyze a huge number of X-ray images and diagnose
patients with the right decision. So, cases of false diagnosis
may occur, where a false diagnosis is defined as either
failure to see a significant fading or attaching the incorrect
diagnosis that is readily seen. A higher false diagnosis rate
will result in weak quality in healthcare and time-delayed
treatment [1].
In 2015, Anu et al [3] extracted features using the Gray-
Level Co-occurrence Matrix (GLCM) from X-ray bone
images. Then, the extracted features were used to calculate
the segmented image and based on these features the bone is
classified whether it has a fracture or not. The presented
methods in [3] work have been tested on a data set with 40
X-ray and CT images containing fractured and non-fractured
(normal) images. In the feature extraction process, the
GLCM features were extracted and the images were
classified depending on that into normal and fractured
images. The performance of their work reached 85%
accuracy. The limitation of the presented method was in
using CT, and some cases of X-ray images, as it was very
problematic to find the area of the fracture [3].
Al-Ayyoub and his colleagues (2013) [4] considered the
binary classification problem to investigate the existence of
the fracture in the hand X-ray images. The dataset consisted
of 98 X-ray images. They focused on Decision Tree (DT),
Bayesian Network (BN), Naive Bayes (NB) and Neural
Networks (NN) methods. Furthermore, as three sets of
features were computed in their work, separate experiments
were conducted to evaluate which set is more useful by
using each set of features individually in the classification
process. But the results were far from perfect and they found
that one way to improve the performance of base classifiers
is to combine all features and use them in classification, also
they tried to use two-level meta-classifiers as it turned out
that it gave the best classification results. In the final stages,
different sets of features were used in the classification
process, but the maximum accuracy level was 91.8% which
was obtained by applying boosting and then bagging on the
Classification of Long Bone X-ray Images using New
features and Support Vector Machine
Amani Al-Ghraibah
1
, Mohammad Algharibeh
2
, Waseem AlMohtasib
3
, Muneera Altayeb
4
1Al-Ahliyya Amman University, Jordan, a.ghraibah@ammanu.edu.jo
2Private Orthopedic Surgery, Jordan, mmgh1990@gmail.com
3Medical Engineering Private Sector, Jordan, was2015eem@gmail.com
4Al-Ahliyya Amman University, Jordan, m.altayeb@ammanu.edu.jo
ISSN 2278
-
3091
Volume 10, No.3, May - June 2021
International Journal of Advanced Trends in Computer Science and Engineering
Available Online at http://www.warse.org/IJATCSE/static/pdf/file/ijatcse011032021.pdf
https://doi.org/10.30534/ijatcse/2021/011032021

Amani Al-Ghraibah et al., International Journal of Advanced Trends in Computer Science and Engineering, 10(3), May – June 2021, 1494 – 1500
1495
Bayesian Network classifiers with the feature set that
included: Wavelets, Curvelets and GLCM features [4].
In Umadevi and Geethalakshmi work [5], twelve features of
two groups; the shape features and texture features were
used. The texture features that were extracted from long
bones X-ray images are Gray Level Co-Occurrence Matrix
(GLCM) features. While the shape features were extracted
using a Fast Hough Transformation algorithm. The accuracy
of a single classification was evaluated using 10-fold cross-
validation technique. Three binary classifiers, SVM, BPNN,
and KNN were presented to build ensemble classification
models. The classifiers were built with different feature sets
and the presented experiments proved that a group of models
can significantly improve the quality of fracture
identification [5].
In 2012, Mahendran and his colleagues [6], focused on their
research to build an automated system that detect fractures in
long bones from diagnostic X-rays using a series of
progressive steps. Three classifiers were considered which
are: Back Propagation Neural Networks, Support Vector
Machine (SVM) and Naïve Bayes. Also, two feature classes
were collected and extracted from X-ray images, namely:
texture and shape features, with a total of 11 features. In the
classification part, four classifiers were merged and used to
classify the X-ray images as fractured or non-fractured
images. The results proved that fusion classifiers were
efficient for fracture detection and reached a maximum
accuracy. One difficulty encountered with fusion
classification was on detecting which classifier produces the
best accuracy. The researchers considered only simple
fractures and the experimental results showed that the
performance reduces if the fractures were parallel or
perpendicular to the bone edge [6].
Mahendran et al [7] used the texture features in bone
fracture detection research. Fused classifiers were proposed
for fracture detection where some specific classifiers were
established and work as a binary classifier, which can report
if a bone fracture is detected or not. If detected, the location
of the fracture is highlighted. There are mainly three
classifiers: Feed Forward Back Propagation Neural
Networks (BPNN), Naïve Bayes (NB) and Support Vector
Machine (SVM) classifier. The fusion classifiers built from
base classifiers which are (1) Texture features with BPNN,
(2) Texture features with SVM, (3) Texture features with
NB and (4) Texture features with BPNN, SVM, and NB.
The basic fusion rule used was that if more than 2 classifiers
report fracture then the image is said to have a fracture.
Many experimental works were accompanied to analyze the
performance of the proposed fusion classifier-based
detection system concerning its efficiency in terms of correct
detection and speed of the algorithm. After comparing the
performance with the traditional single classification system,
the suggested unification of techniques revealed that the
results were improved in terms of accuracy in detecting
fractures and in the speed of the detection [7].
In Myint et al (2018) work [8], suggested a Computer-Aided
Diagnosis (CAD) technique to automatically recognize and
localize the leg bone fracture. Many image processing
techniques were used in their paper, they recognized that
Harris corner detection was an effective tool to catch the
broken points of the leg bone. Decision Tree (DT) was used
as a simple classification for fracture and non-fracture bones.
KNN was also used as it is suitable for pattern recognition
and supports to classify fracture types. Fracture types such
as Transverse, Oblique, Comminuted and Normal were
classified by the system too. The performance accuracy
concerning fracture and non-fracture were calculated and the
accuracy assessment was also evaluated. Kappa accuracy
assessment was used to consider the error results when
calculating the performance and classifying the types of
fractures. However, the system produces the output results
with accurate and reliable performance and less processing
time based on the contributed methods. According to the
result, best accuracy achieved was 83 % using the Kappa
accuracy assessment [8].
In this paper, new features were extracted from the long
bone X-ray images that were discussed in our previous work
[2]. An automated predicting system is built here to predict
the existence of the bone fracture automatically and faster
than the regular diagnosis processes. So, the motivations of
our project are to reduce human errors and reduce the time
and effort associated during the bone injury diagnostic
process which is usually done manually by physicians.
Ultimately, this system can be integrated within the software
of the x-ray imaging devices to allow users to produce a
rapid and highly accurate diagnosis while generating the
image. So, a Graphical User Interface (GUI) is designed
which enables the user to perform interactive tasks.
The novelty of this work covers two main things: using a
large number of Tibia bone X-ray images and classifying the
images based on new features which were investigated in
our previous work depends on the physical properties of the
bone images [2]. Also, Graphical User Interface (GUI) is
built to enable the user to perform interactive tasks easily.
Many classification processes are presented depending on
the feature group used each time and a comparison between
the results is also performed.
This paper is presented as follows: Section 2 explains briefly
the methodology of feature extraction and image’s
classification. Section 3 shows the results of the
classification using different feature groups and summarizes
the graphical user interface (GUI) and Section 4 is the
summary of this work.
2. METHODOLOGY
2.1 Feature Extraction
Recent work has analyzed and extracted new features from
Tibia bone X-ray images depending on the physical

Amani Al-Ghraibah et al., International Journal of Advanced Trends in Computer Science and Engineering, 10(3), May – June 2021, 1494 – 1500
1496
properties of the bone which detects the changes of these
features in the presence and absence of a fracture. Al-
Ghraibah et al produced a novel method to examine whether
these features are efficient in detecting bone fracture or not.
They used the X-ray images of both legs of the patient (the
left and right Tibia bones), taking advantage of human body
symmetry, to study the performance of the extracted features
in detecting bone fracture instead of directly using the
classification methods [2]. Here, we extend the work in [2]
and use the most significant features to build an automated
classification system that classify any X-ray Tibia bone
image as a fractured or non-fractured image.
Four different methods and techniques were used to describe
the physical properties of the bones. These methods are: 1.
wavelet analysis, 2. gradient analysis, 3. Principle
Component Analysis (PCA) and 4. bone edge detection
method. The efficacy of the methods was presented, and the
results showed that, depending on the features changes in the
presence of the fracture, the most significant features
extracted from each method were summarized. Here, a brief
description of each method and the most significant features
are presented.
2.1.1 Wavelet features
Two-Dimensional Discrete Wavelet Transform (2D DWT)
was used in image processing as a powerful tool solving to
image analysis, denoising, image segmentation and other.
2D DWT is computed when the original image is convolved
along x and y directions by low pass and high pass filters.
The images obtained are downsampled by half the size of
the original image. The resultant images are convolved again
with high pass and low pass filters. The four sub-band
images generated contain the approximation coefficient
(which contain the maximum information of the image),
horizontal, diagonal, and vertical information of the image
[9,10].
The three detailed images were used to evaluate the energies
of each decomposition level by adding the absolute values of
the wavelet coefficients (the highpass images). Then a sum
of these energies was calculated including the three highpass
images as we are interested in the edge structure. From that,
five energy values were extracted which are related to each
of the five levels of decomposition namely: Energy level 1,
Energy level 2, Energy level 3, Energy level 4, and Energy
level 5. The results show that all the wavelet energies can
detect the existence of bone fracture in a good way and these
features can be used in further bone classification processes
[2].
2.1.2 Gradient features
From the general image processing science, the spatial
gradient is equivalent to the first derivative of the processed
image. Gradient of the image will highlight fragments and
edges that may not be noticeable in the original image. The
image is filtered (convolved) with the known Sobel filters,
Gx and Gy which are given by: Gx = hx*f and Gy=hy*f
respectively, where * is the two-dimensional convolution
operator, h is the filter and f is the image. To abbreviate this
gradient information into single descriptors for each image,
the following features were computed from the resultant
gradients from each image: mean, standard deviation,
maximum, minimum, skewness and kurtosis [10-12].
From the results in [2], a summary was made that the
gradient features were efficient in detecting bone fracture
and could be used in further bone classification processes.
But it was recommended to exclude: the mean, standard
deviation and minimum features, as they were less
significant in detecting the bone fracture. The remaining and
effective features are the maximum, skewness and kurtosis
and will be used in this research.
2.1.3 Bone edge features
Edge detection defines a set of mathematical procedures that
aim to recognize points in a digital image where the
brightness varies sharply or has discontinuities. One of the
second-order derivative operators that is used for edge
detection is the Laplacian edge detector. It is from the zero-
crossing category of the edge detection technique and it
gives better edge detection results than the first-order
derivative detection techniques, but it is somehow sensitive
to noise [13]. In the previous work [2], the Tibia bone X-ray
images were preprocessed using a smoothing (average) filter
to remove the unwanted signals, the image then was
converted to a binary image and the canny edge detection
method was used to detect the bone edges. From the
resultant image, two features were extracted, which are
related to the edges of the bone; the sum of the on (white)
pixels and the number of the 8-connected pixels in the
binary image. It was found that both features are effective in
detecting bone fracture and could be used in further bone
classification processes.
2.1.4 Principal component features
From a mathematical view, PCA is a statistical process that
uses an orthogonal transformation to convert a set of
observations of probably correlated variables into a set of
values of linearly uncorrelated variables called Principal
Components (PCs). This transformation is defined in such a
way that the first PC has the largest possible variance, which
measures for as much of the variability in the data as
possible. Then, each succeeding component in turn has the
highest variance possible under the constraint that it
is orthogonal to the preceding components [14]. Six features
were extracted from the highest variance PCA which are:
mean, standard deviation, minimum value, maximum value,
skewness, and kurtosis. These six features can detect the
bone fracture effectively depending on the previous work
results [2].

Amani Al-Ghraibah et al., International Journal of Advanced Trends in Computer Science and Engineering, 10(3), May – June 2021, 1494 – 1500
1497
2.2 Classification
Figure 1: General description of SVM
The purpose of supervised, machine learning is to build a
model that makes predictions based on evidence in the
presence of uncertainty. The machine learns from the results
when adaptive algorithms classify data patterns. The
computer improves its prediction performance when
exposed to more observations. Specifically, a supervised
learning algorithm uses a known set of input data and known
responses of the data (classes) and trains a model to generate
reasonable predictions for the response to new data. In this
work, the input X-ray image that has a fractured bone
addressed class 1 and the image without a fractured bone
addressed class 0. Figure 1 shows a simple description of
SVM.
The whole set of input data can be called a heterogeneous
matrix as the matrix rows are called observations or
instances and each of them includes several measurements
for a subject. Matrix columns are denoted to as predictors or
features and each of them represents a measurement of each
subject [15]. In this research, the observations are the X-ray
images of the Tibia bones where the features of each image
are set in columns. The data matrix contains one row of
features extracted from each image as given in (1).
(1)
where n is the number of features extracted and m is the
number of images.All supervised learning methods start with
an input data matrix. The data were prepared as each row in
the feature matrix represents one observation and each
column represents one variable or predictor. In this step the
features were extracted from each X-ray image in the data
set and arranged in one matrix for each image and called a
data matrix. Each row has ten features related to the ten
features described before.
Cross-validation is a statistical method of calculating and
comparing learning algorithms by dividing data into two
segments: the first segment used to train a model and the
other one is to validate the model. The basic form of cross-
validation is k-fold cross-validation, while other forms of
cross-validation are special cases of k-fold cross-validation
or involve repeated rounds of them [16]. A 10-fold cross-
validation method is used in this work where the MATLAB
Table 1: Confusion matrix
Fractured
Non
-
fractured
Fractured
TP
FP
Non
-
fractured
FN
TN
software completes these steps by randomly partition the
data into 10 sets of equal size and train the SVM classifier
on the remaining nine sets. The previous steps were repeated
10 times and at the end, the system combines generalization
statistics from each fold.
2.3 Performance evaluation
A connection between our university and King Abdullah
University Hospital (KAUH) was settled and the data were
collected from the orthopedic department there. The total
number of Tibia bone images used in this work are 100
images for the evaluation purpose of which 50 are with a
fracture while the rest are normal images. The terms used in
the confusion matrix (shown in Table 1) can briefly be
described as: True Positive (TP): true decisive system
classified as true, True Negative (TN): false event detected
as false, False Positive (FP): the event is false and
discriminated as true and False Negative (FN): true event
classified as false [17].
Also, Accuracy (AC) is defined as the probability that the
classification by the system is correct and it is given by (2)
[20]:
 = +
 + + + ∗100(2)
The Sensitivity (True Positive Rate (TPR)) and Specificity
(True Negative Rate (TNR)) are also calculated from the
confusion matrix using (3), and (4) respectively [20]:
 =
 + (3)
 =
 + (4)
3. RESULTS
In this section, the results are presented for the following
experiments: classification using one feature analysis at a
time, classification using all feature analyses, classification
using two or three feature analyses each time. Also, a
Graphical User Interface (GUI) is presented which was built
to classify any new Tibia bone X-ray image into a fractured
or non-fractured image automatically using any of the
existence features group of the user choice.

Amani Al-Ghraibah et al., International Journal of Advanced Trends in Computer Science and Engineering, 10(3), May – June 2021, 1494 – 1500
1498
3.1 Classification using one feature analysis at a time
Table 2: Confusion matrix using gradient features
Fract
ured
Non
-
fractured
Fractured
45
5
Non
-
fractured
23
27
Table 3: Confusion matrix using wavelet features
Fractured
Non
-
fractured
Fractured
46
4
Non
-
fractured
16
34
The result confusion matrix of gradient analysis is shown in
table 2, where the accuracy (AC) is equal to 72%, TPR is
66.2% and TNR is 84.4%. While the confusion matrix of the
wavelet analysis is presented in table 3 and the resultant AC
is 80%, TPR is 74.2% and TNR is 89.5%. From these
results, the wavelet features are more effective than the
gradient features in detecting the bone fracture.
The confusion matrix results after using the edge features is
presented in table 4, where the calculated accuracy AC is
equal to 72%, TPR is 73.8% and TNR is 67.2%. Also, the
confusion matrix of using the PCA features is shown in table
5; the AC is 71%, TPR is 68.3% and TNR is 77.5%. From
the previous results, the accuracy of detecting the bone
fracture using the wavelet analysis is higher than using the
other features: gradient, edge and PCA features. Table 6
presents a summary of the classification results while using
each analysis alone in the classification process. As
mentioned above the wavelet analysis gives the best
classification accuracy (80%) and it shows the best TPR and
TNR results too.
3.2 Classification using all features
The result confusion matrix using all features is shown in
table 7. Where the accuracy (AC) is equal to 73%. While the
TPR and TNR are equal to 71.7% and 74.5%, respectively.
Table 4: Confusion matrix using edge features
Fractured
Non
-
fractured
Fractured
41
9
Non
-
fractured
19
31
Table 5: Confusion matrix using PCA features
Fractured
Non
-
fractured
Fractured
35
15
Non
-
fractured
14
36
Table 6: Summary of classification results using each analysis
alone
Gradient
Wavelet
Edge
PCA
Accuracy (AC)
72%
80%
72%
71%
TPR
66.2%
74.2%
73.8%
68.3%
TNR
84.4%
89.5%
67.2%
77.5%
Table 7: Confusion matrix using all features
Fractured
Non
-
fractured
Fractured
38
12
Non
-
fractured
15
35
3.3 Classification using group of three or two analyses
Other groups of features are used to build the classification
model and classify the images as a fractured and non-
fractured image. Table 8 shows the features groups which
are used in each process which contains a collection of the
features that are extracted using three analyses out of the
four presented analyses. The reason behind using different
features groups is to find the best collection of the extracted
features that could give better classification accuracy. From
table 8 the results show that the best feature group in
detecting the bone fracture is the group of Wavelet, Edge
and PCA features, which means all features except the
gradient features. While the group who gives a balance
accuracy (AC) with TPR and TNR is the group of Gradient,
Wavelet and PCA features (all features except the edge
features). So, the later feature group can detect the fractured
and non-fractured images with the same accuracy.
Table 9 presents the results while using a group of two
feature analyses one of them is the wavelet analysis. In
general, the resultant accuracies are somehow close to each
other with maximum accuracy (AC) is reached using
wavelet and edge features together with high TNR result ~
90%.
3.4 Graphical User Interface (GUI)
The classification system can be integrated within the
software of the X-ray imaging devices so the users can
diagnose the images quickly and accurately during image
generation. Graphical User Interface (GUI) is a graphical
display in one or more windows containing components, that
enable the user to perform interactive tasks. The reason
behind designing an application using a graphical interface
is to make the system more applicable and friendly interface
[18]. In this work, the designed GUI lets physicians to choo-
Table 8: Summary of classification results using three analysis
each time
Gradient,
Wavelet
& Edge
Gradient
,
Wavelet
& PCA
Gradient,
Edge &
PCA
Wavelet,
Edge &
PCA
Accuracy
(AC)
73%
76%
71%
77%
TPR
68.3 %
76%
69.8%
74.5%
TNR
81.1%
76%
72.3%
80%

Amani Al-Ghraibah et al., International Journal of Advanced Trends in Computer Science and Engineering, 10(3), May – June 2021, 1494 – 1500
1499
Table 9: Summary of classification results using two analysis each
time
-se the feature extraction method and then show the resultant
features after using the selected feature method and finally
classify the image into a fractured or non-fractured. Figure 2
shows an example of classifying an X-ray Tibia bone image
using the wavelet analysis method, the image on the left is
the original image selected randomly from the data, the bar
next to the image shows an options for the available feature
extraction methods where the user can select the type of the
method by click on the method name.
The images next to the feature methods bar represent the
image at the last step of image processing which then used to
extract the wanted features. The last bar on the right is the
features that were extracted and used in the classification
process. The image class (fractured or non-fractured) is
presented at the bottom.
4. CONCLUSION
Recent work of Al-Ghraibah et al has analyzed new features
from Tibia bone X-ray images depending on the physical
properties of the bone which detects the changes of these
features in the presence or absence of a fracture. They
produced a novel method to examine whether the extracted
features are efficient in detecting bone fracture or not. In this
work, an extension to the work in [2] is presented and is
used the most significant features to build an automated
classification system that classifies any X-ray Tibia bone
image as a fractured or non-fractured image. Four different
methods were used, namely: wavelet analysis, gradient
analysis, Principle Component Analysis (PCA) and bone
edge detection method. Classification process was repeated
using each feature analysis group alone, using all feature
analyses groups together and using two or three feature
analyses groups at a time. Also, a Graphical User Interface
(GUI) was presented to classify any new Tibia bone X-ray
image into a fractured or non-fractured image automatically
using any of the existence features group of the user choice.
The maximum accuracy result was reached when the
wavelet features are used alone in the classification process
or if they were used along with the edge features too.
Figure 2: GUI example; classifying the input image using wavelet
analysis.
REFERENCES
1. N. Umadeviand S. N. Geethalakshmi. Multiple
classification system for fracture detection in human
bone x-ray images. In 2012 Third International
Conference on Computing, Communication and
Networking Technologies (ICCCNT'12), IEEE, 2012,
pp. 1-8.
2. A. Al-Ghraibah, M. Algharaibeh, W. Al-Muhtaseb, F.
Al-Khateeb, and I. Al-Anis. Investigating the
Significance of New Features Extracted From Long
Bones X-ray Images. 2020IEEE 5th Middle East and
Africa Conference on Biomedical Engineering
(MECBME) rank. IEEE, 2020.
3. T. C. Anu and R. Raman. Detection of bone fracture
using image processing methods. International
Journal of computer applications, 975, pp. 8887, 2015.
4. M. Al-Ayyoub, I. Hmeidi and H. Rababah. Detecting
Hand Bone Fractures in X-Ray Images. J. Multim.
Process. Technol., vol. 4, no. 3, pp. 155-168, 2013.
5. N. Umadevi, and S. N. Geethalakshmi. .Multiple
classification system for fracture detection in human
bone x-ray images. In 2012 Third International
Conference on Computing, Communication and
Networking Technologies (ICCCNT'12), IEEE, July
2012, pp. 1-8.
6. S. K. Mahendran and S. S. Baboo. Ensemble Systems
for Automatic Fracture Detection. International
Journal of Engineering and Technology, vol. 4, no.1,
pp. 7, 2012.
7. S. K. Mahendran and S. S.Baboo. An enhanced tibia
fracture detection tool using image processing and
classification fusion techniques in X-ray
images. Global Journal of Computer Science and
Technology, vol. 11, no.14, pp. 22-28, 2011.
8. W. W. Myint, K. S. Tun and H. M. Tun. Analysis on
Leg Bone fracture detection and classification using
X-Ray images. Machine Learning Research, vol. 3, no.
3, pp. 49-59, 2018.
9. R. C. Gonzalez and R. E. Woods. Image
processing. Digital image processing, 2nd ed.,2007, ch.
7.
Gradient &
Wavelet
Edge
&
Wavelet
PCA &
Wavelet
Accuracy
(AC)
77%
80%
77%
TPR
70.7%
74.2%
73.7%
TNR
88.5%
89.5%
81.4%

Amani Al-Ghraibah et al., International Journal of Advanced Trends in Computer Science and Engineering, 10(3), May – June 2021, 1494 – 1500
1500
10. A. Al-Ghraibah, L. E.Boucheron and R. T. J.McAteer.
An automated classification approach to ranking
photospheric proxies of magnetic energy build-
up. Astronomy & Astrophysics, vol. 579, no. A64, 2015.
11. R. C. Gonzalez, S. L. Eddins and R. E. Woods. Digital
image publishing using MATLAB. Prentice Hall.
2004.
12. S. Gupta and S. G. Mazumdar. Sobel edge detection
algorithm. International journal of computer science
and management Research, vol. 2, no. 2, pp. 1578-
1583, 2013.
13. R. Anas, H. A. Elhadi and E. S. Ali. Impact of Edge
Detection Algorithms in Medical Image
Processing. World Scientific News, vol. 118, pp. 129-
143, 2019.
14. I. T. Jolliffe and J. Cadima. Principal component
analysis: a review and recent
developments. Philosophical Transactions of the Royal
Society A: Mathematical, Physical and Engineering
Sciences, vol. 374, no. 2065, pp. 20150202, 2016.
15. W. L. Chao. Machine learning tutorial. Digital Image
and Signal Processing, 2011.
16. P. Refaeilzadeh, L. Tang and H. Liu. Cross-
Validation. Encyclopedia of database systems, vol. 5,
pp. 532-538, 2009.
17. A. Tharwat. Classification assessment
methods. Applied Computing and Informatics. Vol.
ahead-of-print, No. ahead-of-print, 2020.
18. M. Kalechman.. Practical MATLAB basics for
engineers. Crc Press, 2018.