Identification of Glioma using Discrete Wavelet Transform (DWT)

and Artificial Neural Network (ANN)

Endah Nur Salamah

, Dian C Rini Novitasari

, Ahmad Hanif Asyhar

, Muh Ma’arif

Program Studi Matematika, Universitas Islam Negeri Sunan Ampel, Surabaya, Indonesia

Keywords: Glioma, Discrete Wavelet Transform (DWT), Artificial Neural Network (ANN)

Abstract: Glioma is one of the deadly diseases and suffered by many people in the world. Glioma means brain tumor.

In 2016, the World Health Organization (WHO) recorded as many as 6.2 million people of the world suffering

gliomas. Based on this fact, it is necessary to examine glioma using a tool that is one of magnetic resonance

imaging (MRI), then the results of brain MRI image be analyzed or diagnosed by an expert doctor but

sometimes the results of its analysis is still subjective and takes a long time. The image used in this study are

the normal brain MRI image and glioma brain MRI image. First steps are image improvement (Adaptive

Histogram Equalization), the second step is image segmentation using the Otsu threshold, third step is image

extraction using discrete wavelet transform (DWT) with features taken are energy, standard deviation and

mean, then classification using ANN (backpropagation) which will be identified into two classes namely

normal and glioma. Based on the testing result using the Matlab program, the results of image extraction using

the best decomposition levels DWT Haar is 4. Then the results of the best-hidden layers backpropagation

ANN classification is 25 then obtained MSE error value = 0,0000999 it indicates that the model used in this

study is suitable for identifying gliomas using image data and obtained the best accuracy values of 91.67%,

the sensitivity of 100%, and specificity of 85.71%.

1 INTRODUCTION

There are many kinds of brain disease, one of them is

a tumor. Many people in the world suffered the

disease. Uncontrolled growth of body cells that

changes character, shape and kinetic is called a tumor

(Lestari, et al., 2017). In this study, we identified the

tumor that occurs in glia cells is called glioma

(Drevelegas, 2011). Glia cells in the brain are neuron

supporting cells that decisive the synaptic contact and

protect the ability of neurons signals so the signal

connection of neurons and brain properly worked

(Purves, et al., 2001). World Health Organization

(WHO) recorded that 688,000 people in the United

States suffered the primary glioma in 2012 and 6.2

million of the world's population suffered the glioma

in 2016 (Anitha & Raja, 2017). Glioma is the fifth

disease that causes the death of women cancer

patients at age 20-39 years (Sari, et al., 2013). Data

cases of glioma in Indonesia haven’t reported, so

there are just a few data and no records of glioma data.

Glioma patients in Indonesia increased in 2013,

especially the Gatot Soebroto Army Hospital in

Jakarta (Satyanegara, 2010). The test of brain tumors

is taken by specialists using CT-Scan or MRI devices

which the results will analyze by a doctor, but

sometimes the results of doctor's diagnosis are

subjective and require a long time. Therefore, a

method is needed to help the diagnosis be more

objective in a short time.

In this study used brain MRI images data because

the results are a clear image and almost 90% used

MRI to diagnose the head and spinal cord (Sjahriar &

Iwan, 2005). Anitha and Raja obtained the features of

brain tumors for identification of brain cells in normal

and tumor glia cells in their paper "Segmentation of

Glioma Tumors using Convolutional Neural

Networks" using morphological operations, there are

opening and closing to segmented the features and

background in brain images (Anitha & Raja, 2017).

Neural networks are used in many problems such as

prediction, EEG signal classification, and image

classification. Classification of EEG signal used

Adaptive Neighborhood Modified Backpropagation

Salamah, E., Novitasari, D., Asyhar, A. and Ma’arif, M.

Identiﬁcation of Glioma using Discrete Wavelet Transform (DWT) and Artiﬁcial Neural Network (ANN).

DOI: 10.5220/0008906700002481

In Proceedings of the Built Environment, Science and Technology International Conference (BEST ICON 2018), pages 289-294

ISBN: 978-989-758-414-5

289

(ANMBP) with EEG signal data in normal and

epilepsy patients (Novitasari, 2015). The first study

referred by the author of the study is backpropagation

neural network implementation for medical image

compression which applies backpropagation method

for processing medical images with X-ray

examination of bones into four classes there are

fractured, dislocated, broken and healthy with

accuracy values obtained by 90% (Dimililer, 2013).

The second study referred is Guijarro et all in discrete

wavelet transform for improving greenness image

segmentation in agricultural image by extracting the

features of agricultural image using DWT texture

analysis to obtained value of mean and standard

deviation, so image quality is obtained with good

results although the image that used is a low-quality

image (Guijarro, et al., 2015).

A related study that made by the author based on

the previous study is about the identification of

glioma using discrete wavelet transform (DWT) and

artificial neural network (ANN) to identify and

classify brain MRI images into two classes, there are

normal brain and glioma. In this study used

segmentation process to separate objects and

background from brain images. DWT is the

technique of low filters and high filters as image

processing used to image extraction to obtain the

value of energy, mean, and standard deviations, it also

to find which the DWT process that obtained the best

decomposition level value of glioma texture. The

three values used as input to the ANN method with a

backpropagation algorithm. In this study used the

backpropagation algorithm because the previous

research obtained high accuracy values at the process

of medical image data. Backpropagation algorithm

used to find the proper classification process with the

best backpropagation structure.

2 LITERATURE REVIEW

2.1 Glioma

Glioma is a type of disease that occurs in tissues that

grow uncontrollably and abnormally in the brain. The

brain has an important role in regulating body activity

because the brain is the central neuron system. The

main cause of gliomas is still unknown. There are

several factors that can increase a person's risk of

developing a glioma including hereditary factors and

side effects of radiotherapy (radiation) procedures.

Gliomas that arise from glia cells are called gliomas

and this type of brain tumor is most common. This

type of glioma is located supratentorially above the

tentorium of the brain (Enggariani, 2012). In this

study two classes were used: normal and glioma

presented in Figure 1. Identification of glioma

visually has a different or striking color. The image

used in this study results from Magnetic Resonance

Imaging (MRI) examination.

Figure 1: Brain MRI image (Anitha & Raja, 2017)

2.2 Discrete Wavelet Transform

(DWT)

Wavelet transformation is the transformation of

discrete signals into wavelet coefficients obtained by

filtering signals using two filters, namely a high pass

filter and low pass filter. Mathematical operations

that are the main basis of wavelets include (Sutarno,

2010):

1) Translation, example 



  



 



  







  



2) Scaling, example 







 







  







Based on the two mathematical operations above that

combined produce a family of wavelets. Wavelets

have many types depending on the functions used,

such as Haar, Symlet, Daubechies, Coifflet, and so

on. The most commonly used types of wavelets are

Haar and Daubechies.

DWT is a transformation that is identical to the

hierarchical sub-band system where the logarithmic

sub-band is in the frequency domain. DWT

decomposes the image of the four sub-bands of

frequency, namely the low-pass through low pass

filter (LL), the low pass through high filter (LH), the

high pass through low pass filter (HL), and the high

pass through high pass filter (HH) such as in Figure 2

(Nayak, et al., 2015). The filter on the image is called

decomposition.

Figure 2: The level 1 wavelet decomposition

(Guijarro, et al., 2015)

A The Brain Normal

B Glioma

BEST ICON 2018 - Built Environment, Science and Technology International Conference 2018

290

DWT Haar is used in this study because it is better

to represent the characteristics of texture and shape

and the computation time needed is less compared to

other wavelet methods (Novamizanti & Kurnia,

2015). The analysis of the texture that is used such as

energy, average, and standard deviation (Singh, et al.,

2015). Energy is used to calculate Euclidean distance

which is then used to determine similarity and the

inequality of two feature vectors, mean and standard

deviation are statistical groups whose purpose is to

find out the uniformity and concentration of a data in

an image, then the feature vector is used for input

artificial neural networks.

2.3 Artificial Neural Network (ANN)

The artificial neural network is one information

processing system that is designed by imitating the

work of the neuron system in the human brain. ANN

consists of two layers, namely the input layer and the

output layer which has a different number of

networks (neurons), but there are have layers located

between the input and output layers called hidden

layers (Novitasari, et al., 2016). In this study, the

algorithm used is the backpropagation algorithm

because based on the previous study obtained high

accuracy values and good for classification of image

data (Dimililer, 2013). Backpropagation algorithm is

a supervised algorithm that trains the network by

spreading the output error backward from the output

layer to the input layer. The backpropagation training

algorithm using the sigmoid activation function is as

follows (Siang, 2005):

a. Initialize all weights with small random

numbers.

b. If the termination condition has not been

met, do step c-j.

c. Each input unit receives a signal and passes

it to a hidden unit above it.

d. Calculates all outputs in hidden units which

is the sum of input signals that have been

given weight and bias.

e. Calculates all network outputs in the output

unit.

f. Calculate the error factor of the output unit

based on errors in each output unit that

serves to correct errors and determine

changes in weight to be used.

g. Calculate the error factor of the output unit

based on errors in each network unit (node)

output, then calculate the change in weight.

Next, calculates all changes in weight from

the output unit to the hidden unit.

2.4 Measurement of Classification

Accuracy

In the classification results using backpropagation

algorithm, it is necessary to have a system validation

to ensure its accuracy, so that the confusion matrix is

used by analyzing the value of sensitivity, specificity,

and accuracy (Akobeng, 2006). The confusion matrix

structure is presented in Figure 3.

Figure 3: Structure Confusion Matrix

Sensitivity is a test to detect the right ill patient is

diagnosed with the illness. The method is the TP

value divided by the TP value plus the FN value

multiplied by 100%. Specificity is a test to detect that

a normal patient is diagnosed as normal. The method

is the TN value divided by the TN value plus the FP

value multiplied by 100%. Accuracy is a test to

correctly identify the actual condition of the patient

(positive or negative results that are right). The

method is to divide the exact number of classification

with the total data multiplied bay 100%.

3 RESEARCH METHODS

3.1 Data

In this study, the data used was obtained from the

Atlas web. Atlas web is a website that provides

information and data in the form of images that are

specifically the brain, such as normal brain image

data, glioma image data, brain cancer images, etc.

(Johnson & Becker, 1995). This study uses data as

many as 60 images consisting of the normal brain and

tumor images with a comparison of training and

testing data that is equal to 80% and 20% and 75%

and 25% (Hota, et al., 2013).

3.2 Preprocessing

The first step is the brain image data is converted to

grayscale image (gray), then cropping is done, then

gray intensity is improved by the adaptive histogram

equalization process, then the image is refined using

the median filter. This study will be carried out in

Identiﬁcation of Glioma using Discrete Wavelet Transform (DWT) and Artiﬁcial Neural Network (ANN)

291

several main processes is preprocessing,

segmentation, extraction, and classification with

brain MRI image data. The process is shown in Figure

Figure 4: The classification system of glioma image

3.3 Segmentation

Next step, after preprocessing finish is the

segmentation process, which is the separation of

objects with the background using the Otsu threshold

method.

3.4 Extraction

In this process, it functions to take the pattern of

normal brain image and glioma pattern using the

Discrete Wavelet Transform Haar method then

analyzed the pattern texture in the image by

calculating the energy, average, and standard

deviation values which are then used as input for the

classification process ANN backpropagation.

3.5 Classifications

The fourth step is the image classification process.

This process uses backpropagation method for data

training/testing. The image data used is 60 images

consisting of normal brain and glioma. Comparison

of training and testing data used is 75%, 25% and

80%, 20%. After classification, the results were tested

using a confusion matrix by analyzing the value of

sensitivity, specificity, and accuracy. Furthermore, it

classified into 2 classes, namely normal or tumor. The

structure of the backpropagation model used in this

study is in Figure 5.

Figure 5: The structure backpropagation of study data

(Prasetyo, 2014)

4 RESULTS AND ANALYSIS

The results of the trial of brain image processing and

classification using the MatLab program.

4.1 Preprocessing

The results all processing steps use the MatLab

presented in Figure 6. Based on the preprocessing

results that have been done provide a level of clarity

of the glioma object, so the preprocessing results can

be applied at the segmentation step.

Figure 6: The results of preprocessing brain image

Input Layer

Hidden Layer

Output

Layer

BEST ICON 2018 - Built Environment, Science and Technology International Conference 2018

292

4.2 Segmentation

At this steps, the function in MatLab is used, such as

level = graythres (I) so the results are presented in

Figure 7. Based on the results of the segmentation

obtained the color difference between the object and

the background image that the object is at the level of

intensity of 0 and 1 pixels, where 0 is the pixel

intensity in black and 1 is the color of white pixels.

Figure 7: The results of the segmentation otsu threshold

brain image

4.3 Extraction

In this study used the image extraction process using

Discrete Wavelet Transform Haar method with the

decomposition level used is 4 and 5. The results of

image extraction are presented in Figures 8 and 9.

Figure 8: The results of level 4 DWT Haar brain image

Figure 9: The results of level 5 DWT Haar brain image

At level 4 there are 13 low and high filters and the

trial process needs a bit of time to reach the error

target. At level 5 there are 16 low and high filters and

the trial process takes quite a bit of time to reach the

error target. In this feature extraction process

obtained vector DWT Haar feature with different

decomposition levels, namely level 1-5 which is sized

according to the level used. Suppose that the

decomposition level used is 3 then got the result of

feature vector DWT with size 1 × 30 pixels which will

be used as input to the Artificial Neural Network

(ANN) Backpropagation.

4.4 Classifications

Backpropagation classification in this study was

carried out computationally using Matlab software.

Then the training and testing processes are shown in

Figure 10.

Figure 10: The Results of the Training and Testing

Backpropagation

Based on Figure 10 the best results of

backpropagation process for glioma identification are

80% training data and 20% testing data using 4

decomposition levels whose results are vectors values

of energy, mean, and standard deviations of 1 × 39

lengths for each image, then the best backpropagation

structure is 25 nodes of the hidden layer, learning rate

equal to 0,00070, and MSE error value is 0.0000999.

Then a comparison is made to check the

classification results in the backpropagation model

with the results of the actual classification. The result

is that there is a normal brain image classified as

glioma so the specificity value is 85.71%, the

sensitivity value is 100%, and the accuracy value is

91.67%.

5 CONCLUSIONS

Based on the testing result using the Matlab program, the

results of image extraction using the best decomposition

levels DWT Haar is 4. Then the results of the best-hidden

layers backpropagation ANN classification is 25 then

obtained MSE error value = 0,0000999 it indicates that

the model used in this study is suitable for identifying

gliomas using image data and obtained the best

accuracy values of 91.67%, the sensitivity of 100%, and

specificity of 85.71%.

Identiﬁcation of Glioma using Discrete Wavelet Transform (DWT) and Artiﬁcial Neural Network (ANN)

293

REFERENCES

Akobeng, A., 2006. Understanding Diagnostic Test1:

Sensitivity, Specify, and Predictive Values.

Jurnal Acta Pediatrica, pp. 338-341.

Anitha, R. & Raja, D. S. S., 2017. Segmentation of

Glioma Tumors using Convolutional Neural

Networks. International Journal Imaging

System Technology, pp. 354-360.

Dimililer, K., 2013. Backpropagation Neural

Network Implementation for Medical Image

Compression. Journal of Applied

Mathematics, pp. 1-7.

Drevelegas, A., 2011. Imaging of Brain Tumor with

Histological Correlations. second ed. New

York: Springer.

Enggariani, 2012. Tumor Otak (Brain Tumors),

Pekanbaru: Fakultas Kedokteran Universitas

Riau.

Guijarro, M., Riomorros, I., Pajares, G. & Zitinski, P.,

2015. Discrete Wavelet Transform for

Improving Greeness Image Segmentation in

Agricultural Images. Computer and

Electronics in Agriculture, pp. 396-407.

Hota, H. S., Shrivas, A. K. & Singhai, K. S., 2013.

Artificial Neural Network, Decission Tree

and Statistical Techniques Applied for

Designing and Developping E-mail

Classifier. International Journal of Recent

Technology and Engineering (IJRTE), pp.

164-169.

Johnson, K. A. & Becker, J. A., 1995. Retrievied from

the Whole Brain Atlas. [Online]

Available at:

http://www.med.harvard.edu/aanlib/

[Accessed Juli 01 2018].

Lestari, Y., Mesran, Suginam & Fadlina, 2017.

Sistem Pakar untuk Mendiagnosis Penyakit

Tumor Otak menggunakan Metode

Certainty Factor (CF). Jurnal INFOTEK,

Volume 2 , pp. 82-86.

Nayak, D. R., Dash, R. & Majhi, B., 2015. Brain MRI

Image Classification using Two-

dimensional Discrete Wavelet Transform

and AdaBoost with Random Forest. Jurnal

Neurocomputing, pp. 1-10.

Novamizanti, L. & Kurnia, A., 2015. Analisis

Perbandingan Kompresi Haar Wavelet

Transform dengan Embedded Zerotree

Wavelet pada Citra. Jurnal ELKOMIKA, pp.

161-176.

Novitasari, D. C. R., 2015. Klasifikasi Sinyal EEG

menggunakan Metode Fuzzy C-Means

Clustering (FCM) dan Adaptive

Neighborhood Modified Backpropagation

(ANMBP). Jurnal Matematika "MANTIK",

pp. 31-36.

Novitasari, D. C. R., Farida, Y. & Puspitasari, 2016.

Klasifikasi menggunakan Metode Hybrid

Bayessian-Neural Network (Studi Kasus:

Identifikasi Virus Komputer). Jurnal

Matematika "MANTIK", pp. 38-43.

Prasetyo, E., 2014. Data Mining- Mengolah Data

menjadi Informasi menggunakan Matlab.

Yogyakarta: ANDI.

Purves, D., Augustine, G. J. & Fitzpatrick, D., 2001.

Neuroscience. second ed. Sunderland:

Sinauer Associates.

Sari, E., Windarti, I. & Wahyuni, A., 2013.

Karakteristik Klinis dan Histopatologi

Tumor Otak di Dua Rumah Sakit Bandar

Lampung, Lampung: Fakultas Kedokteran

Universitas Lampung.

Satyanegara, 2010. Ilmu Bedah Saraf. 4 ed. Jakarta:

Gramedia Pustaka Utama.

Siang, J. J., 2005. Jaringan Saraf Tiruan &

Pemrogramnya menggunakan Matlab.

Yogyakarta: Andi Offset.

Singh, A. et al., 2015. Image Processing Based

Automatic Diagnosis of Glaucoma using

Wavelet Features of Segmented Optic Disc

from Fundus Image. Computer Methods and

Programs in Biomedicine, pp. 108-120.

Sjahriar, R. & Iwan, E., 2005. Radilogi Diagnostik ,

Jakarta: Fakultas Kedokteran Universitas

Indonesia.

Sutarno, 2010. Analisis Perbandingan Transformasi

Wavelet pada Pengenalan Citra Wajah.

Jurnal GENERIC, pp. 15-21.

BEST ICON 2018 - Built Environment, Science and Technology International Conference 2018

294