Available online at www.sciencedirect.

com

ScienceDirect
Procedia Computer Science 102 (2016) 39 – 44

12th International Conference on Application of Fuzzy Systems and Soft Computing, ICAFS 2016, 29-30
August 2016, Vienna, Austria

Effect of image enhancement on MRI brain images with neural networks

Kamil Dimililera*, Ahmet ølhanb
a,*
Department of Electrical and Electronic Engineering, Near East University, P.O.Box:99138, Nicosia, North Cyprus, Mersin 10 Turkey
b
Computer Engineering Department, Near East University, P.O.Box:99138, Nicosia, North Cyprus, Mersin 10 Turkey

Abstract

Since human body’s standard control metabolism stops, old cells do not die and these abnormal cells forms a mass of tissue, known as tumor.
In order to diagnose tumors, huge biomedical machines such as MRI, PET-CT, and MG machines are used. Last decade, there are many studies
in brain tumor detection in magnetic resonance imaging (MRI). Location of the brain tumor and precise size are detected with brain tumor
detection. In this paper, image processing techniques are applied on MRI images to preserve information details. These techniques are erosion,
contrast enhancement and median filtering. The purpose of observe is develop an image processing algorithm for brain cancer detection on
MRI images. Comparison of back propagation neural networks will be done using original images and reconstructed images on the effect of
categorization.
© 2016
2016 The
The Authors.
Authors.Published
PublishedbybyElsevier
ElsevierB.V.
B.V. This is an open access article under the CC BY-NC-ND license
Peer-review under responsibility of the Organizing Committee of ICAFS 2016.
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the Organizing Committee of ICAFS 2016
Keywords: Contrast Enhacement; MRI; Brain Tumor; Artificial Neural Networks

1. Introduction

In literature, brain tumors are defined as malignant cells rise up in the brain. These cancerous cells grow into a mass of cancer
tissues that interferes with brain capabilities consisting of muscle control, sensation, memory, and different everyday body
capabilities2. As stated in WHO's report in early 2000s, fatal injuries by cancer is 6.2 million in worldwide.
Most cancerous cells are referred to as malignant tumors, and people composed of specifically non-cancerous cells are called
benign tumors. There are two types of brain tumors named as primary and secondary. Cancer cells that build from brain tissue are
known as primary brain tumors, whereas tumors that spread from different organs to the brain are grouped as secondary brain
tumors or metastatic. Brain tumors are categorized as a four malignant stages which is called Grade I, II, III and IV2, 3.
Benign brain tumor that consist cancer cells, can be removed. Benign brain tumors usually have an explicit side or edge. They
do not expand to other parts of the body. However, benign tumors can induce serious health problems. Grade I and II are benign
brain tumors3, 4, 5. The other type of tumors, which is called as malignant brain tumor consists of cancer cells and it is also called
as brain cancer. They are likely to grown up swiftly and can affect nearby normal brain tissues. This type of tumors can be a
threat for life. Malignant brain tumors are in grade III and IV3, 4, 5.
In diagnosis and early detection of tumors, medical imaging plays an important role. Magnetic resonance imaging (MRI),
computed tomography (CT), and different imaging techniques provide an efficient way to detect diseases6. MRI uses a magnetic
subject and pulses of radio waves to make images of organs and systems within the body. In many cases, MRI offers different

* Corresponding author. Tel.: +90 (392) 444 0 983 / 380; fax: +09 (392) 223 64 61.
E-mail address: kamil.dimililer@neu.edu.tr

1877-0509 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the Organizing Committee of ICAFS 2016
doi:10.1016/j.procs.2016.09.367
40 Kamil Dimililer and Ahmet İlhan / Procedia Computer Science 102 (2016) 39 – 44

facts about systems in the body than may be visible with a MG, or CT scan. MRI may also display problems that cannot be
visible with different imaging methods6, 7.
Researchers are using multi-disciplinary approach comprising knowledge in medicine, mathematics and computer science to
get better understanding the disease and applied treatment way effectively. Since brain images are complicated and tumors can
be analysed only by expert physicians, medical image processing for brain tumor detection is one of the challenging tasks in
computer science1.
Various applications include digital image processing such as detection of criminal face, figure print authentication system,
object recognition etc. Brain tumor detection which is an application of medical imaging plays an important role. Brain tumor
detection depends on the affected part of the brain along with its shape, size, and boundary14. For such duties, image processing
techniques were used with the aid of artificial neural networks. Back propagation is a supervised system and popular in learning
techniques with the multi-layer network. In back propagation neural networks, process of information flows from the direction of
the input layer towards output. The learning is achieved by adjusting the connection weight in back propagation neural networks
iteratively so that trained. Iteration numbers of the training algorithm and the convergence time will vary depending on diagnosis
problem with their various data set15.
In this research two phases are designed. These phases are called Image with Neural Networks (IWNN), and Image
Processing with Neural Networks (IPWNN). The aim of this paper is to perform image processing techniques, as well as
artificial neural networks and their interrelated analysis methods to benign brain tumors patients. Ultimate research relay on
quantitative information for instance, the shape, size and the ratio of the affected cells17, 18.

2. Systems

In this paper, an Image with Neural Networks (IWNN), and Image Processing with Neural Networks (IPWNN) are designed.
The suggested system is an image processing model that combines different techniques in the field of image processing20. The
goal of this system is to detect brain tumors located within the MRI image. Matlab programming language is used for
implementing and simulating the suggested system (Matlab 2013a).
Initially, morphological operations have been applied to the input image. Afterwards, the original input image is firstly
processed by morphological operations and then with image erosion technique. Contrast enhancement and median filtering are
applied to the processed image respectively.
The suggested system will be implemented using 50 images that are randomly selected within the medical image database.
The images are of size 256*256 raw type images.

3. Proposed Systems

This section focusses on utilities and parameters of the proposed system. IWNN is a backpropagation neural network which
uses the original input brain image set, whereas IPWNN is a backpropagation neural network system that uses image processing
techniques applied to the original images.

3.1. IWNN: Image with Neural Networks

In this system, two phases are designed. First phase is preparation of the input image for the neural network phase which is
considered as second phase. Figure 1 represents the sample images from the database.

(a) (b)
Fig. 1. Original Sample images
 Kamil Dimililer and Ahmet İlhan / Procedia Computer Science 102 (2016) 39 – 44 41

3.1.1. Data Preparation Phase

Initially, the original image has been converted to grayscale in order to minimize the CPU time. The grey color image contain
of pixel intensity between 0-255 where 0 represents black and 255 is for white10. Original image sizes are 256 x 256, whereas
64 x 64 images are preferred considering the processing time. These two resizing can be found in the figure 2.

(a) (b)

Fig. 2. (a) 256x256 Original Image; (b) 64x64 Resized Image

3.1.2. Neural Networks Phase

IWNN uses a conventional 3-layer back propagation neural network with 4096 input neurons, 35 hidden neurons and 2 output
neurons classifying the organs with tumors and without tumors. Output neurons classify the organs using binary coding: [1 0]
for the Organ with Tumor and [0 1] for the Organ without Tumor. The sigmoid activation function is used for activating neurons
in the hidden and the output layers. Figure 3 represents the Neural Network topology and Table 1 represents the final parameters
of IWNN.

2 1
2
IWNN

Reduced Size 35
Original Image Image (64x64) 4096
(256x256) pixels pixels Received Image

Fig. 3. Neural Network Topology of IWNN.

Table 1: Trained neural network final parameters IWNN

Input Layer Nodes 4096

Hidden Layer Nodes 35
Output Layer Nodes 2
Learning Rate 0.004
Momentum Rate 0.5
Minimum Error 0.002
Iterations 2713
Training Time 36 seconds
42 Kamil Dimililer and Ahmet İlhan / Procedia Computer Science 102 (2016) 39 – 44

3.2. IPWNN: Image Processing with Neural Networks

IPWNN uses a conventional 3-layer back propagation neural network with 4096 input neurons, 16 hidden neurons and 2
output neurons classifying the organs with tumors and without tumors. Output neurons classify the organs using binary coding:
[1 0] for the Organ with Tumor and [0 1] for the Organ without Tumor. The sigmoid activation function is used for activating
neurons in the hidden and the output layers. Figure 4 represents the Neural Network topology and Table 2 represents the final
parameters of IPWNN.

2 1
2
IPWN
N
2

Reduced Size 16
& 4096
Original Image Received Image
(256x256) Adjusted
pixels Image
(64x64)
pixels
Fig. 4. Neural Network Topology of IPWNN.

Table 2: Trained neural network final parameters IPWNN

Input Layer Nodes 4096

Hidden Layer Nodes 16
Output Layer Nodes 2
Learning Rate 0.005
Momentum Rate 0.6
Minimum Error 0.002
Iterations 3676
Training Time 45 seconds

4. Discussions

Figure 3 and figure 4 shows the topologies of the suggested BPNN Systems. Due to the implementation simplicity, and the
availability of sufficient “input – target” database for training, the use of a back propagation neural network which is a
supervised learner, has been preferred.
Training and testing (generalization) are comprised for these phases. The available brain image database is organized as
follows:
1. Training image set: 20 images
2. Testing image set: 30 images
During the learning phase, initial random weights of values between -0.4 and 0.4 were used. In order to achieve the required
minimum error value and meaningful learning, the learning rate and the momentum rate were adjusted during various
experiments. An error value of 0.002 was considered as sufficient for this application.

5. Results

These results were obtained using a 2012R2 64 bits OS Virtual machine that has a 1 Gbits connection speed to a super
computer that has IBM Blade HS21 XM Intel Xeon 2.33 GHz CPU with 1280 core units. Through this application, the
robustness, flexibility and speed of this intelligent brain tumor detection system have been demonstrated. Tumor identification
results using the training image set yielded 100% recognition as would be expected in both BPNN.
The testing image sets of IWNN and IPWNN results were successful and encouraging. An overall correct identification of
IWNN yielded 83% correct identification where, 25 images out of the available 30 brain images yielded. An overall correct
 Kamil Dimililer and Ahmet İlhan / Procedia Computer Science 102 (2016) 39 – 44 43

identification of IPWNN yielded 90% correct identification where, 27 images out of the available 30 brain images yielded. This
successful result was obtained by using only the database of images for training the neural network.

(a) (b)

Fig.5. (a) IWNN RMS Error Graph; (b) IPWNN RMS Error Graph
44 Kamil Dimililer and Ahmet İlhan / Procedia Computer Science 102 (2016) 39 – 44

As it can be seen from the results of the iterations using the Super computer, IWNN finalized the training after 2713 iterations
within 36 seconds, whereas IPWNN finalized the training process after 3676 iterations within 45 seconds. Additionally, hidden
layer nodes, learning rates, and momentum rates for both IWNN and IPWNN have been obtained after several experiments.
Figure 5 (a) and Figure 5 (b) shows the RMS Error graphs for the two neural networks.
IPWNN results using the testing image sets were successful and encouraging. An overall correct identification of IPWNN
yielded 90% correct identification where, 27 images out of the available 30 brain images yielded. These successful results were
obtained by using only the database of images for training the neural network.

6. Conclusions

In this paper, two new approaches for the detection of brain tumors were developed. These new approaches are based on
image processing techniques as well as neural networks. The used image processing techniques are very useful and significant in
the medical field such as in the MRI machine.
The robustness in the developed system is to stimulate the human visual inspection that usually highlights or notice the
presence of any abnormality in the brain directly by the first look. Likewise, the system marks the abnormalities or tumors in an
iris directly in 0.02 seconds which makes our system robust and effective.
The images used in testing the proposed system are collected from two databases to get a total of 50 images: 25 normal and 25
abnormal.
The experimental results obtained when testing the proposed system proved that the developed brain tumor detection system
is a robust image processing system that is capable of detecting any abnormalities in the brain region particularly at small
patches.

References

1. Noureen E, Hassan K. Brain Tumor Detection Using Histogram Thresholding to Get the Threshold point. IOSR Journal of Electrical and Electronics
Engineering 2014; 9(5): 14-9.
2. Masalkar D N, Shitole A S. Advance Method for Brain Tumor Classification. International Journal on Recent and Innovation Trends in Computing and
Communication 2014; 2(5): 1255-9.
3. Sharma Y, Chhabra M. An Improved Automatic Brain Tumor Detection System. International Journal of Advanced Research in Computer Science and
Software Engineering 2015; 5(4): 11-5.
4. Mansa S M, Kulkarni N J, Randive S N. Review on Brain Tumor Detection and Segmentation Techniques. International Journal of Computer Applications
2014; 95: 34-8
5. Anandgaonkar G, Sable G. Brain Tumor Detection and Identification from T1 Post Contrast MR Images Using Cluster Based Segmentation, International
Journal of Science and Research 2014; 3(4): 814-7.
6. Ravi A, Sreejith S. A Review on Brain Tumor Detection Using Image Segmentation. International Journal of Emerging Technology and Advanced
Engineering 2015; 5(6): 60-4.
7. Mustaqeem A, Javed A, Fatima T. An Efficient Brain Tumor Detection Algorithm Using Watershed and Thresholding Based Segmentation. I. J. Image
Graphics and Signal Processing 2012; 10(5): 34-9.
8. Saini P K, Singh M. BRAIN Tumor Detection in Medical Imaging Using Matlab. International Journal of Engineering and Technology 2015; 2(2): 191-6.
9. Swathi P S, Devassy D, Vince P, Sankaranarayanan P N. Brain Tumor Detection and Classification Using Histogram Thresholding and ANN. International
Journal of Computer Science and Information Technologies 2015; 6(1): 173-6.
10. Ghare S, Gaikwad N, Kulkarni N, Nerkar M. Detection of Possibility of Brain Tumor Using Image Segmentation. International Journal of Innovative
Research in Computer and Communication Engineering 2015; 3(4): 2701-4.
11. Joseph R P, Singh C S, Manikandan M. Brain Tumor MRI Image Segmentation and Detection in Image Processing. International Journal of Research in
Engineering and Technology 2014; 3(1): 1-5.
12. Sachin N, Shah D, Khairnar V, Kadu S. Brain Tumor Detection Based on Bilateral Symmetry Information. International Journal of Engineering Research
and Applications 2014; 4(6): 98-103.
13. Sravanthi K, Samiuddin S. Brain Tumor Detection, Demarcation and Quantification via MRI. International Journal & Magazine of Engineering
Technology, Management and Research 2015; 2(8): 672-6.
14. Selkar R G, Thakare M N. Brain Tumor Detection and Segmentation by Using Thresholding and Watershed Algorithm. International Journal of
Advanced Information and Communication Technology 2014; 1(3): 321-4.
15. Vally D, Sarma Ch V. Diagnosis Chest Diseases Using Neural Network and Genetic Hybrid Algorithm. International Journal of Engineering
Research and Applications 2015; 5(1): 20-6.
16. Logeswari T, Karnan M. An improved implementation of brain tumor detection using segmentation based on soft computing. Journal of Cancer
Research and Experimental Oncology 2010; 2(1): 6-14.
17. Dimililer K. Neural Network Implementation for Image Compression of X-Rays. Electronics World 2012; 118(1911): 26-9
18. Dimililer K. Back Propagation Neural Network Implementation for Medical Image Compression.Journal of Applied Mathematics 2013;453098:1-8.
19. Khashman A, Dimililer K. Comparison Criteria for Optimum Image Compression. The International Conference on Computer as a Tool
(EUROCON’05), Serbia & Montenegro, 21-24 November 2005.
20. Dimililer K, Kirsal-Ever Y, Ugur B. Tumor Detection on CT Lung Images using Image Enhancement. International Science and Technology
Conference (ISTEC 2016), Austria, 13-15 July 2016.