Proceedings of the Second International Conference on Inventive Research in Computing Applications (ICIRCA-2020)

IEEE Xplore Part Number: CFP20N67-ART; ISBN: 978-1-7281-5374-2

Detection of Bone Fractures Automatically with

Enhanced Performance with Better Combination of
Filtering and Neural Networks

Syed Karimunnisa Pradeep Raj Savarapu

Department Of CSE, Department Of CSE,
Koneru Lakshmaiah Education Foundation, Koneru Lakshmaiah Education Foundation,
Vaddeswaram,AP,India.522502. Vaddeswaram,AP,India.522502.

Ram Kumar Madupu CMAK Zeelan basha

Department Of CSE,
Department Of CSE,
Koneru Lakshmaiah Education Foundation
Koneru Lakshmaiah Education Foundation,
,Vaddeswaram,AP,India.522502.
Vaddeswaram,AP,India.522502.
P.Neelakanteswara
Department Of CSE,
Koneru Lakshmaiah Education Foundation
,Vaddeswaram,AP,India.522502

Abstract—Major parts affecting the accidents are bone and Keywords—BPNN(Back Propagation Neural
severe bone fractures lead to the death of human beings. Timely Network),ANN(Artificial Neural Network)Harris corner
treatment is to be provided to the injuries, and sometimes detection,SVM(Support Vector Machine),MRI,CT Scan,X-Rays.
orthopedic surgeons need to take immediate decisions regarding
surgeries to the injuries. Among the available X-Ray, MRI, and
CT scan options for bone fracture detection, X-Rays are mostly
used due to its availability with less cost to all kinds of people. I. INTRO DUCTIO N
Here an automatic X-ray fractures detection technique is
proposed which helps orthopaedecians to take an immediate According to the reports of WHO (World Health Organisation)
decision for surgeries and to detect the fracture even without the reports the rate of accidents is increasing year to year. And in
support of an orthopaedician. In this paper, a back propagation most of the accidents, death rates are 30-40% and in cases like
neural network (BPNN) along with canny edge segmentation and fractures in the abdomen, the death rate is near to 100%.In
a conservative smoothing filtering technique is proposed which
helps in getting a better accuracy rate when compared to S VM medical imaging bone fracture detection is a crucial part.
(S upport Vector Machine) and ANN (Artificial Neural Network). Sometimes it is very difficult to identify Hairline fractures due
The proposed methodology also compares with the Harris corner to low quality of x-ray images. And long day work of
detection technique. Among all other techniques a combination of
radiologists and orthopaedics also leads to missing of detection
BPNN with Canny edge detection and conservative smoothing
proves to better performer for automatic bone fracture detection some fracture in some cases. Also in many places where there
with accuracy of 91%. is deficiency of doctors it is very difficult to provide timely
treatments for the injuries. In order to overcome this CAD
(Computer Aided Diagnosis) is a solution, which reduces the
. burden on orthopaedicians.

978-1-7281-5374-2/20/$31.00 ©2020 IEEE 189

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 26,2023 at 13:40:46 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Inventive Research in Computing Applications (ICIRCA-2020)
IEEE Xplore Part Number: CFP20N67-ART; ISBN: 978-1-7281-5374-2

[1]There are many proposals for Computer Aided Diagnosis segmentation of long bones from digital x-rays is introduced in
where the accuracy of detecting the fractures in a bone are less. [6].Data is collected after inspecting various bone shapes. To
Here an efficient bone fracture detection is proposed with identify the bone in an image, after performing edge detection,
which applied methodologies helps in getting better accuracy an edge which belongs to boundary should be in the calculated
rate. Here in proposed work a dataset 900 X-ray images are ranges. Scale and rotation invariant algorithm is used. Partial or
taken from Andhra Hospitals, Vijayawada, out of which 500 complete bones can be identified. In [7] brief introduction of
are fractured and 400 are normal images. computer-aided diagnosis (CAD) is explained briefly. It has
become one of the important research in medical imaging and
In this work a combination of Conservative Filtering, Canny radiology. Various Cad schemes are explained. To improve the
Edge Detection, and Texture Feature Extraction is used to use of CT images in preoperative planning [8], a 3D modeling
enhance the efficiency in detection of fractures.Fig.1 represents and simulation surgery. In computer-aided preoperative
the architecture of proposed work. In the architecture is shown surgery, bone segmentation and reduction are two significant
initially input x-ray image is given as input to the system. For simulation tools.
input image pre-processing is done in order to remove noise in
the images. [2][3]Next stage comes is the segmentation where
edges are found. From then texture features are extracted. III. P ROPOSED METHODOLOGY
Which are given as input to the neural network for training. Fig.2 represents the clear steps for the proposed methodology.
At the first step an x-ray image is give as input. [9]This X-ray
image is converted to the gray scale image and its PSNR value
is checked where conservative filtering proves to be best when
compared to other filtering techniques which is proves by its
higher PSNR value for all images. Then filtered image is fed
for segmentation where canny edge detection is applied which
is resulting best among Sobel and perwitt. [10]From the
segmented image texture features are extracted like mean,
median. These texture features are given as input to the
different neural networks like SVM, ANN, BPNN.

Fig.1 Architecture of Proposed System

II. LITERATURE SURVEY

An important problem in clinical orthopedics is determining

the power of a healing fracture. The feasibility of computing
neural networks for determining the vibration of broken bone
strength has been demonstrated using computer simulations
[4].In [5], it is stated that the bone electric conductivity differs
Fig.2 Proposed methodology
from the radial direction to the direction parallel to the bone
axis. It suggests the bone has the behavior of anisotropic
electrical. The generated e.m.f in the bone varies with the stress
that is applied on the fractured bone. An algorithm for

978-1-7281-5374-2/20/$31.00 ©2020 IEEE 190

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 26,2023 at 13:40:46 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Inventive Research in Computing Applications (ICIRCA-2020)
IEEE Xplore Part Number: CFP20N67-ART; ISBN: 978-1-7281-5374-2

3.2 Segmentation
3.1 Filtering Segmentation is a phase which helps in extracting the image
Filtering is used for improving the quality of an image. edges. Segmentation is a technique that divides the image into
Filtering is applied in order to reduce the noise from the the parts based upon the intensity levels of an image. The
image. For this PSNR plays a major role in understanding the main motive behind applying segmentation techniques is to
noise in the image. More PSNR is the indication of the least extract the features if the region of interest in an image. The
noise. So before applying the filtering technique, image PSNR pre-processed image is fed as input to the image for
value is checked and if the PSNR value of any image is less in segmentation. [12]The reason behind applying the canny edge
the dataset then it is must undergo the pre-processing detection is its clearness in getting the region of interest. [13]If
technique. Conservative filtering works better in this regard. the image is segmented correctly then it is very easy to extract
For removing the noise in an image here a special technique the features of image properly. By input image having
that preserves the high-frequency components and also which intensity levels one is tissue level second is bone and the third
performs the filtering by using the minimum and the is a background in which we need only bone intensities which
maximum value of the neighborhood pixels in an image is helps much in extracting the features. Fig.5 is a pre-
smoothing filtering. [11]This method is specialized to remove processed image and the fig.6 is segmented image with canny
the noise from the image. i,e: the pixels with uncommon edge detection.
intensities and whose pixels are isolated (salt and pepper
noise)... It uses a 3x3 neighbourhood for the original image.
The filter ranks the values of neighbourhood pixels of the
original image i.e., leaving the pixel in the second row-third
column which is the target pixel, and compares if the pixel
value is greater than the maximum value in the
neighbourhood. It replaces this pixel value with the maximum
value if it is greater than the maximum value in the
neighbourhood otherwise it compares with the minimum
Fig.5 Filtered Image Fig.6 Segmented Image
value. It replaces with the minimum value if that pixel value is
less than the minimum neighbourhood. This results in the
3.3 Feature Extraction of Texture features:
noiseless image which can be used for the further
Texture features of the bone such as shape, and orientation of
process.Fig.3 and Fig.4 are input image and the filtered image.
a bone features are extracted. [14][15]Texture features
extraction is done in order to give the inputs to the neural
network. Texture features of the edges are taken as mean and
median values which are fed as input with which neural
network gets trained and further helps in detecting the
fractures in the image.
3.4 Classification using Neural Network
[16]Obtained texture features are fed as input to the neural
network for training purpose. Neural network consists of three
Fig.3 Input image Fig.4 Filtered Image layers, they are input layer, hidden layer and the output layer.

978-1-7281-5374-2/20/$31.00 ©2020 IEEE 191

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 26,2023 at 13:40:46 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Inventive Research in Computing Applications (ICIRCA-2020)
IEEE Xplore Part Number: CFP20N67-ART; ISBN: 978-1-7281-5374-2

Here in the proposed work three kinds of neural networks are class Accuracy Sensitivity Specificity
taken. [17]They are SVM, BPNN, ANN.Fig.7 represents the SVM 0.86 0.83 0.84
neural network architecture.
BPNN 0.91 0.88 0.86

ANN 0.88 0.85 0.8

Table.1 Comparison of Performance

Table.1 shows the accuracy, sensitivity, specificity values of
SVM,BPNN,ANN.Fig.8 is statistical report of performance
evaluation of different specifiers .

Fig.7 Neural Network Architecture

IV.RESULTS
Proposed work is done with Matlab 2019b.Where out of 500
Normal images 300 images are used for testing and 200 are
used for testing and out of 400 abnormal images 200 images
used for training and 200 images are used for testing.Fig.7
Fig.8 Representation of results
shows the result of bone fracture detection automatically.
Filtered Image in fig.4 which gets obtained as filtered one is V.CONCLUSION
given with features after segmentation for training the neural In the proposed work 500 images which are normal and 400
network. The result is tested with three neural networks SVM, are abnormal. Here from 500 normal, 300 are taken for testing
and 200 for training and from 400 abnormal images, 200 for
BPNN, ANN where SVM classifier gives the accuracy of testing, and 200 for training. Results are compared using 3
86%, BPNN gives the accuracy of 0.91 and ANN gives kinds of classifiers, like ANN, BPNN, and SVM in which
BPNN gives a better classification of 86% when compared to
0.88.Which is represented in Table.1. the other 2 classifiers as it has better combination of filtering
technique. Future Scope of the proposed work is, this work
can be extended for detecting the fractures in a curved bones
also.

REFERENCES
[1] Selvikvåg Lundervold, A., & Lundervold, A. (2018). An overview of
deep learning in medical imaging focusing on MRI. Zeitschrift Für
Medizinische Physik. doi:10.1016/j.zemedi.2018.11.002
[2] Ibrahim, W. H., Osman, A. A. A., & Mohamed, Y. I. (2013). MRI brain
image classification using neural networks. 2013 INT ERNAT IONAL
CONFERENCE ON COMPUT ING, ELECT RICAL AND
ELECT RONIC ENGINEERING (ICCEEE).
doi:10.1109/icceee.2013.6633943
Fig.7 Fracture Detection [3] Khemphila, A., & Boonjing, V. (2011). Heart Disease Classification
Using Neural Network and Feature Selection. 2011 21st International
Conference on Systems Engineering. doi:10.1109/icseng.2011.80
[4] C. Saha and M. F. Hossain, "MRI brain tumor images classification
using K-means clustering, NSCT and SVM," 2017 4th IEEE Uttar

978-1-7281-5374-2/20/$31.00 ©2020 IEEE 192

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 26,2023 at 13:40:46 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Second International Conference on Inventive Research in Computing Applications (ICIRCA-2020)
IEEE Xplore Part Number: CFP20N67-ART; ISBN: 978-1-7281-5374-2

Pradesh Section International Conference on Electrical, Computer and [11] Basha, C. Z., Krishna, A., & Savarapu, P. R. (2019). Automatic
Electronics (UPCON), Mathura, 2017, pp. 329 -333. detection of lung infection. International Journal of Recent T echnology
[5] S. Chaplot, L.M. Patnaik, N.R. Jagannathan, "Classification of magnetic and Engineering, 8(3), 200–203.
resonance brain images using wavelets as input to support vector [12] Jaskirat Kaur, Sunil Agrawal, Renu Wig, “A comparative analysis of
machine and neural machine", Biomedical Signal Processing and thresholding and edge detection segmentation techniques.” International
Control, vol. 1, no. 1, pp. 86-92, January 2006. journal of computer applications. Vol. 39, No. 15, pp. 29 –34, 2012.
[6] Basha, C. Z., Sricharan, K. M., Dheeraj, C. K., & Ramya Sri, R. [13] S.K. Mahendran, “A comparative study on edge detection algorithms
(2018). A study on wavelet transform using image analysis. International for computer aided fracture detection systems”. International journal of
Journal of Engineering and T echnology(UAE), 7(2), 94–96. engineering and innovative technology. Vol. 2, No. 5, pp. 191 –193,
[7] Zeelan Basha, C. M. A. K., Maruthi Padmaja, T., & Balaji, G. N. (2018). November 2012.
Automatic X-ray image classification system. In Smart Innovation, [14] Zahari Abu Bakar, Nooritawati Md T ahir and Ihsan MYassin,
Systems and T echnologies (Vol. 78, pp. 43–52). Springer Science and “Classification of Parkinson’s Disease Based on Multilayer Perceptron
Business Media Deutschland GmbH. Neural Network”, 2010 6thInternational Colloquium on Signal
[8] C.M A.K. Zeelan Basha, Maruthi Padmaja, and G.N.Balaji,“Computer Processing & ItsApplications (CSPA).
Aided Fracture Detection System” Journal of Medical Imaging and [15] Mahmoud Al-ayoub, Ismail Hmeidi, Haya Rababah, “Detecting hand
Health Informatics Vol. 8, 526–531, 2018 . bone fractures in X-ray images”, Journal of multimedia processing and
[9] Zeelan Basha, C. M. A. K., Maruthi Padmaja, T., & Balaji, G. N, “EAI technologies, Vol. 4, No. 3, pp. 155–168, September 2013.
Endorsed T ransactions on Pervasive Health and T echnology”, Vol. 5 [16] Rafeal. C. Gonzalez & Woods, “Chapter 10: Image segmentation”,
,Issue 18, 1-6,2019.. Digital image processing, Pearson Education, Inc., 2005, pp. 392 –439.
[10] C. Z. Basha, M. R. K. Reddy, K. H. S. Nikhil, P. S. M. Venkatesh and [17] Vijayakumar, T . (2019). COMPARAT IVE ST UDY OF CAPSULE
A. V. Asish, "Enhanced Computer Aided Bone Fracture Detection NEURAL NET WORK IN VARIOUS APPLICAT IONS. Journal of
Employing X-Ray Images by Harris Corner T echnique," 2020 Fourth Artificial Intelligence, 1(01), 19-27.
International Conference on Computing Methodologies and
Communication (ICCMC), Erode, India, 2020, pp. 991 -995.

978-1-7281-5374-2/20/$31.00 ©2020 IEEE 193

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY CALICUT. Downloaded on October 26,2023 at 13:40:46 UTC from IEEE Xplore. Restrictions apply.