Identification of Alzheimer’s Disease in MRI Data using Discrete

Wavelet Transform and Support Vector Machine

Putri Wulandari, Dian Candra Rini Novitasari and Ahmad Hanif Asyhar

Departement of Mathematics, UIN Sunan Ampel Surabaya

Jl Ahmad Yani No. 117 Surabaya, Indonesia

Keywords: Support Vector Machine, Discrete Wavelet Transform, Alzheimer’s Disease.

Abstract: Dementia is a serious problem, recorded worldwide as 4.6 million cases of dementia each year, 60-70%

caused by Alzheimer’s disease. Alzheimer’s disease interferes with daily activities that can lead to death. In

order to obtain proper treatment by a specialist, early detection is required. So, this paper aims to assist the

medical in diagnosing Alzheimer’s disease. Detection of Alzheimer’s disease begins with segmentation the

feature of magnetic resonance imaging (MRI) data using Fuzzy C-Means (FCM ) into three clusters, using

Discrete Wavelet Transform (DWT) to extract the features of all sub-band ‘Haar’, ‘Daubechies 2’, and

‘Daubechies 4’, and classified using the Support Vector Machine (SVM) into two classes: Alzheimer and

non-Alzheimer. The result shows that approximation sub-band third level wavelet transformations in ‘Haar’

is the best method to identify Alzheimer’s disease, with the accuracy value is 97.37%, the sensitivity value

to detect Alzheimer’s disease is 100%, and the specificity value is 92.86%.

1 INTRODUCTION

Most of the elderly people degenerated central

nervous system. It caused a progressive loss of

cognitive function called dementia. It is a serious

problem, 4.6 million cases of dementia each year,

60-70% caused by Alzheimer’s disease was

recorded (Alzheimer’s Disease International, 2008).

Alzheimer’s symptoms are characterized by memory

impairment, changes in mood and personality,

problematic interactions, and abstract thinking (Al-

Naami, 2013). Memory and cognitive of

Alzheimer’s Patients will be decreased for 3 to 9

years. Alzheimer’s disease interferes with daily

activities that can lead to death. Predicted in the next

20 years, people with Alzheimer’s will increase year

by year (Zhang et al., 2011). In order to obtain

proper treatment by a specialist, early detection is

required. It is also possible for Alzheimer’s patients

to plan future decisions before their condition

becomes worse and they considered a burden to their

environment.

Non-invasive methods used to diagnose or

observe the brain tissue of people with Alzheimer’s

are magnetic resonance imaging (MRI) and positron

emission tomography (PET). Compared by other

imaging techniques, MRI is the best choice because

the images result of the anatomical structures of the

brain with brain tissues different contrasted, fewer

artifacts, faster and without using X-ray radiation

(Nayak et al., 2016). The test results of Alzheimer’s

patients MRI looks the abnormalities in the cortical

and periventricular areas, there are hippocampal

atrophy and amygdala in the subcortical region, as

well as enlargement of the basal cisterna and fissure

Sylvia which is early symptoms of dementia (The

National Academy on an Aging Society, 2000).

GM WM CSF

Figure 1: Brain Tissues.

In Figure 1, the brain has three basic tissue

classes, there is gray matter (GM), white matter

(WM), and cerebrospinal fluid (CSF). David Wilson

used GM segmentation results to reduce errors in

classification because statistical analysis of GM can

reduce false positive findings (Wilson &

198

Wulandari, P., Novitasari, D. and Asyhar, A.

Identiﬁcation of Alzheimer’s Disease in MRI Data using Discrete Wavelet Transform and Support Vector Machine.

DOI: 10.5220/0008519301980204

In Proceedings of the International Conference on Mathematics and Islam (ICMIs 2018), pages 198-204

ISBN: 978-989-758-407-7

Laxminarayan, 2007). One of a common technique

used to segmentation image is Fuzzy C-Means

(FCM), the method classified the data into multiple

classes by assign the members of data to the center

of the cluster (Afifah, Rini, & Lubab, 2016). Iraki

Khalifa, et. al. (2012) segmented MRI brain image

using a combined algorithm called Wavelet Fuzzy

C-Means (WFCM). He used the Wavelet method for

feature extraction and FCM to segment into three

classes.

Wavelet transformation is one of the common

image analysis techniques used to extract features. It

gives many feature space, also a good time and

resolution to generated wavelet coefficient with

strong features that can improve the accuracy in

classification (Aiswarya & Simon, 2013). Luis

Javier H, et. al. using Discrete Wavelet Transform

(DWT) at extract features, Principal Component

Analysis (PCA) at reducing features and NMIRS at

features selection to identify Alzheimer’s disease in

Mild Cognitive Impairment (MCI) conditions. The

results show that dimensional reduction in PCA and

NMIRS processes can cause the results of the

classification have poor accuracy and preferably use

the SVM method to obtain better accuracy (Herrera

et al., 2013). Lahmiri & Boukadoum (2013)

analyzed MRI data using multiscale analysis (MSA)

to get fractals with six different scales using a

Support Vector Machine (SVM). It gives the results

from 93 classified MRI brain data; 51 images are

normal brains and 42 images are Alzheimer’s.

In this paper, we identified Alzheimer’s disease

based on MRI data using FCM to segment the GM

characteristics of the brain. Furthermore, DWT is

used to extract the statistical data of the

segmentation reduction brain, and classified into two

categories, Alzheimer or non-Alzheimer, using

SVM.

2 MANUSCRIPT PREPARATION

2.1 Alzheimer

Alzheimer’s is one of the causes of dementia, which

causes memory loss and progressive personality

changes (Al-Naami, 2013). Alzheimer’s disease was

first discovered by Alois Alzheimer’s when

examining an elderly patient who was confused and

difficult to understand questions and had a chaotic

memory. Based on the stages of Alzheimer’s

disease, there are preclinical, mild cognitive

impairment, and dementia stages. Alzheimer’s

disease begins when ‘plaque’ proteins are between

nerve cells and damage to the nerve fibers area.

Patients with Alzheimer’s need special care, because

patients will have severe memory problems,

confusion, and difficulty understanding questions,

such as time, places, pictures, situations, and others

(Mareeswari et al., 2015).

2.2 Histogram Equalization

To improve the image that the pixel distribution is

uneven (having a range of distant values) is used

histogram equalization (Kaur, 2015). Histogram

Equalization produces an image output whose pixel

intensity over a dynamic range is evenly distributed

(Pandey et al., 2016). Histogram Equalization can be

expressed in the transformation function in

Equation(1):

T (x) =











Maksimum Intensity

(1)

where N is the total value of pixels in the image and

is the pixel value at the intensity i.

2.3 Fuzzy C-Means

In 1973, Dunn the first time demonstrated FCM

which was further refined by Professor Jim Bezdek

in 1981 (Janani et al., 2013). FCM is part of Fuzzy

Clustering which is used to analyze patterns of data

(Febrianti et al., 2016). From the results of the

analysis, the data is processed to be grouped,

segmented, or classified. In Fuzzy Clustering, each

data point has a degree of the cluster so that cluster

edge points will be clustered to a lower level than

the cluster center.

To obtain the result of segmentation, the first

step by representing the frequency value of image

data. Then create a vector from minimal to maximal

from the data and select a random central point with

a minimum value is 2. After that calculate the

membership matrix and cluster center. Then the

process stops if the condition has been fulfilled

(Mohammed et al., 2016).

2.4 Discrete Wavelet Transform

Wavelet is a mathematical function used to describe

data into different frequency components, and it will

be studied each component according to its scale

resolution (Herrera et al., 2013). There are many

types of wavelet families, but the type frequently

used is Haar and Daubechies. At each level, it will

pass through high-pass and low-pass filter processes

(Novitasari, 2015). Discrete Wavelet Transformation

Identiﬁcation of Alzheimer’s Disease in MRI Data using Discrete Wavelet Transform and Support Vector Machine

199

(DWT) 2D assumes wavelet coefficients in four sub-

band images, they are LL, LH, HL, HH (Janani et

al., 2013). It represents four decomposition layers

that shown in Figure 2, the component detail is the

horizontal direction for LH (Low-High), vertical for

HL (High-Low), diagonal for HH (High-High) and

Approximation for LL (Low-Low) (Nayak et al.,

2016).

Figure 2: Wavelet Decomposition.

The information is provided by DWT can be

used to statistical analysis and signal synthesis (Isar,

Moga & Lurton, 2005). In DWT 2D the information

is used to get features values of images, they are the

value of the mean, standard deviation, and entropy.

Mean =















(2)

Std =

















n - 1



(3)

Entropy =





i,j



(-ln P



i,j



)

i,j



(4)

2.5 Support Vector Machine

SVM is one of the learning methods used to the

detection of classification, regression, and outliers

(Evgeniou & Pontil, 2001). The idea of

implementing SVM is the value of a vector is

mapped into a high-dimensional feature space. The

SVM method has two basic steps, they are training

and testing. The value of accuracy, specificity, and

sensitivity need to be known to test the accuracy of

classification of data testing based on previous

training.

Accuracy =

TP+TN

TP + TN + FP+ FN



(5)

Specificity =

TN+ FP



(6)

Sensitivity =

TP+ FN

(7)

The Accuracy value is the value that measures

the success rate of the classification performed.

Sensitivity is a value that measures how many

people who have the disease are correctly diagnosed

that it is diseased. Meanwhile, the specificity value

is the inverse of sensitivity value,is a value that

measures how many normal people are correctly

diagnosed that it isn’t diseased (Nayak et al., 2016).

To determine the value of True Positive (TP), True

Negative (TN), False Positive (FP) and False

Negative (FN) can be seen in Table 1.

Table 1: Confusion Matrix.

Actual

Classification

3 METHODS

In this paper, this type of research is included into

applicative research because the input and output

data for the identification of Alzheimer’s disease

using DWT and SVM is numerical data, which

results analysis aims to assist the medical in

diagnosing Alzheimer’s disease.

The data used are MRI brain axial data obtained

from Alzheimer’s Disease Neuroimaging Initiative

(ADNI) and E-Health Laboratory. The concentration

of this research is to classify MRI brain data into

two categories: Alzheimer’s or non-Alzheimers. The

proposed method through four steps, they are pre-

processing step before being processed in the next

step, feature segmentation using FCM, feature

extraction using DWT and Binary SVM for

classification. For the process scheme in detail

described Figure 3.

ICMIs 2018 - International Conference on Mathematics and Islam

200

Figure 3: Diagram of the Process

The first step, pre-processing is applied to ease

image processing, sometimes the owned image data

has a different contrast, lighting, and noise each

image data. In the pre-processing step, MRI brain

data input is grayscale images. Furthermore, for the

normalized image then image applied enhancement

by histogram equalization.

Furthermore, the second step is using FCM, to

obtain feature segmentation of GM, WM, and CSF,

the number of clusters which inserted is three. From

the three features of GM, WM, and CSF, we just

used the GM feature. In feature extraction, we used a

third level decomposition wavelet of ‘Haar’,

‘Daubechies 2’, and ‘Daubechies 4’. The coefficient

values of each sub-band, we used the value of the

mean, standard deviation, and entropy as features

vector that input into the SVM classification step.

In the classification step, the data divided into

two using K-Fold Cross-validation, there are training

data and testing data with the ratio is 60:40. The

statistical value of four features, mean, standard

deviation and entropy of each coefficient sub-band

‘Haar’, ‘Daubechies 2’, and ‘Daubechies 4’ from

training data is used to obtain optimal SVM model.

4 RESULTS AND DISCUSSION

The identification of Alzheimer’s disease using

several methods. They are FCM used to

segmentation, DWT used to extraction and SVM

techniques used to a classification of brain MRI

data. Before feature extraction is done to get features

used as entered SVM Classify, pre-processing and

segmentation step is required. The pre-processing

and segmentation results are represented in figure 3.

The next, the features are taken from GM image

on each coefficient sub-band (approximation,

horizontal, vertical, diagonal) such as value of mean,

standard deviation, and entropy in ‘Haar’,

‘Daubechies 2’, and ‘Daubechies 4’ (as shown in

Table 2) used as input in SVM classification. Then

to validate the value of accuracy, specificity, and

sensitivity classified result used Confusion Matrix.

( a ) ( b) ( c ) ( d ) ( e )

Figure 4: Image Processing: (a) Brain MRI, (b) Histogram Equalization, (c) GM, (d) CSF, (e) WM.

Identiﬁcation of Alzheimer’s Disease in MRI Data using Discrete Wavelet Transform and Support Vector Machine

201

Table 2: Sample Data of ‘Haar’, ‘Daubechies 2’, and ‘Daubechies 4’ Feature Extraction Results.

DWT

Haar

daubechies 2

daubechies 4

Mean

Standard

Deviation

Entropy

Mean

Standard

Deviation

Entropy

Mean

Standard

Deviation

Entropy

Approximation

Coefficient

0.92790

1.06963

-8525.1

0.89383

1.12633

-8959.7

0.77470

1.12954

-10743.1

1.05097

1.04035

-9784.4

1.01031

1.10373

-10139.4

0.86936

1.11494

-12076.5

0.85547

1.00409

-10208.4

0.82263

1.05912

-10679.6

0.70922

1.05904

-12830.5

0.91542

1.05529

-10505.7

0.87999

1.12865

-10978.0

0.75765

1.12752

-13124.1

1.00198

1.10337

-9355.2

0.96331

1.15995

-9986.5

0.82974

1.16532

-12100.5

Horizontal

Coefficient

-0.00644

0.40793

-14573.1

0.00294

0.40136

-31989.2

-0.00611

0.37279

-36245.0

-0.00618

0.43333

-14170.3

0.00292

0.44677

-45716.6

-0.00568

0.42042

-49755.0

0.00234

0.44067

-13842.5

-0.00471

0.43706

-45176.4

0.00054

0.42566

-49888.2

0.01574

0.44483

-14349.4

-0.00348

0.46092

-43948.3

0.01570

0.44368

-51455.3

-0.01178

0.44307

-13150.9

0.00301

0.43729

-44058.6

-0.01057

0.39870

-50625.2

Vertical

Coefficient

-0.00290

0.43375

-14213.3

0.00525

0.40736

-31700.8

-0.00321

0.38676

-35541.2

0.00410

0.44014

-14297.1

-0.00392

0.44948

-45879.4

0.00240

0.42577

-47522.9

0.00265

0.41185

-13870.4

-0.00808

0.42242

-45852.7

0.00021

0.39422

-49069.0

-0.00120

0.44549

-14361.0

-0.01616

0.40899

-16339.7

-0.00839

0.38546

-48696.2

-0.01239

0.43696

-13063.2

0.01290

0.41123

-43632.0

-0.00665

0.38254

-50172.9

Diagonal

Coefficient

-0.00075

0.21045

-15760.9

0.00079

0.21856

-48450.7

0.00144

0.21133

-50337.4

-0.00074

0.24626

-15927.6

-0.00259

0.24379

-76079.1

-0.00023

0.24867

-73724.7

0.00094

0.24736

-14871.2

0.00258

0.23843

-74795.1

0.00095

0.23325

-75186.2

-0.00036

0.22875

-15631.7

0.00024

0.22668

-18111.1

-0.00106

0.22462

-72641.2

-0.01007

0.22859

-14288.6

-0.00896

0.21956

-72062.9

-0.00468

0.21597

-74847.9

Based analysis of MRI data features in each sub-

band in ‘Haar’, ‘Daubechies 2’, and ‘Daubechies 4’

which become input binary SVM Classifier,

obtained different percentage value of accuracy,

sensitivity, and specificity.

Table 3: Value of Accuracy, Sensitivity, and Specificity.

Wavelet Transform

Accuracy

Sensitivity

Specificity

Daubechies 4

Approximation

97.37

100.00

92.86

Horizontal

92.11

100.00

78.57

Vertical

89.47

100.00

71.43

Diagonal

92.11

100.00

78.57

Daubechies

Approximation

97.37

100.00

92.86

Horizontal

86.84

100.00

64.29

Vertical

92.11

100.00

78.57

Diagonal

89.47

91.67

85.71

Haar

Approximation

97.37

100.00

92.86

Horizontal

94.74

100.00

85.71

Vertical

94.74

100.00

85.71

Diagonal

97.37

100.00

92.86

Based on table 3 we can see that the selection of

(approximation) sub-band wavelet

transformations in ‘Haar’, ‘Daubechies 2’, and

‘Daubechies 4’ has the same result, with the

accuracy value is 97.37%, the sensitivity value to

detects Alzheimer’s disease is 100%, and the

specificity value is 92.86%. Based on 38 Testing

data, 24 data are correctly identified as Alzheimer’s

and 13 data as Non-Alzheimer. Meanwhile, one of

non-Alzheimer’s data was identified as Alzheimer’s,

shown in Table 4. However, if viewed based on all

sub-band on each wavelet family tested, ‘Haar’ is

the best solution.

5 CONCLUSION

Dementia is a serious problem that 3 million cases of

Alzheimer’s disease were recorded. Alzheimer’s

disease interfered daily activities that can lead to

death. We identified Alzheimer’s disease based on

MRI data using FCM to segment and DWT to

extract the GM characteristic features of the brain.

Furthermore, we used SVM to classify into

Alzheimer or non-Alzheimer categories based on the

analysis of GM MRI data features. Each sub-band

(approximation, horizontal, vertical, diagonal) in

‘Haar’, ‘Daubechies 2’, and ‘Daubechies 4' obtained

a different percentage of accuracy, sensitivity, and

specificity values. So, we concluded that using

approximation sub-band third level wavelet

transformations in ‘haar is the best solution to

identify Alzheimer's disease, with the accuracy value

is 97.37%, the sensitivity value to detects

ICMIs 2018 - International Conference on Mathematics and Islam

202

Alzheimer’s disease is 100%, and the specificity

value is 92.86%.

Table 4: Result of SVM Classification using ‘Haar’

Approximation Third Level Wavelet Transform.

Data Classification

Binary SVM Classification

Alzheimer

Non-Alzheimer

Alzheimer

REFERENCES

Afifah, N., Rini, D. C., & Lubab, A., 2016. Pengklasteran

Lahan Sawah di Indonesia sebagai Evaluasi

Ketersediaan Produksi Pangan Menggunakan Fuzzy

C-Means. Mantik, 02(01), 40–45.

Aiswarya, V. S., & Simon, J., 2013. Diagnosis of

Alzheimer’s disease in Brain Images using Pulse

Coupled Neural Network. International Journal of

Innovative Technology and Exploring Engineering

(IJITEE), 2(6), 99–101.

Al-Naami, et. al., 2013. Automated Detection of

Alzheimer Disease using Region Growing Technique

and Artificial Neural Network. World Academy of

Scince, Engineering and Technology, International

Journal of Biomedical and Biological Engineering Vol

7(5): 204 – 208. Retrieved from

http://www.waset.org/publications/11271.

Alzheimer’s Disease International, 2008. Global

Alzheimer’s Disease Charter. Retrieved from

https://www.alz.co.uk/global-charter.

Evgeniou, T. & Pontil, M., 2001. Support Vector

Machines: Theory and Applications. Machine

Learning and Its Applications, pp. 249–257, Part of

the Lecture Note in Computer Science book series

(LNCS, Vol. 2049).

Febrianti, F., Hafiyusholeh, Moh. & Asyhar, A. H., 2016.

Perbandingan Pengklusteran Data Iris Menggunakan

Metode K-Means Dan Fuzzy C-Means. Jurnal

Matematika “Mantik,”02(01), 7–13.

Herrera, L. J., Rojas, I., Pomares, H., Guillen, A.,

Valenzuela, O., & Banos, O., 2013. Classification of

MRI Images for Alzheimer’s Disease Detection. 2013

International Conference on Social Computing, 846–

851.

Isar, A., Moga, S., & Lurton, X., 2005. A Statistical

Analysis of the 2D Discrete Wavelet Transform. Proc.

of the International Conference AMSDA, May 17-20,

Brest, France, pp. 1275-1281.

Kaur, S., 2015. Noise Types and Various Removal

Techniques. International Journal of Advanced

Research in Electronics and Communication

Engineering (IJARECE), 4(2), 226-230.

Khalifa, I., Youssif, A., & Youssry, H., 2012. MRI Brain

Image Segmentation based on Wavelet and FCM

Algorithm. International Journal of Computer

Applications, 47(16), 32–39.

Lahmiri, S., & Boukadoum, M., 2013. Alzheimer’s

Disease Detection in Brain Magnetic Resonance

Images using Multiscale Fractal Analysis. ISRN

Radiology, 2013, Article ID. 627303.

http://dx.doi.org/10.5402/2013/627303.

Mareeswari, S., & Dr, G., 2015. A Survey: Early

Detection of Alzheimer’s Disease Using Different

Techniques. International Journal on Computational

Science & Applications, 5(1), 27–37. DOI:

10.5121/ijcsa.2015.5103.

Mohammed, H. R., Alnoamani, H. H., & Jalil, A. A.,

2016. Improved Fuzzy C-Means Algorithm for Image

Segmentation. International Journal of Advanced

Research in Artificial Intelegence, 5(6), 7-10.

Nayak, D. R., Dash, R. and Majhi, B., 2016. Brain MR

Image Classification using Two-Dimensional Discrete

Wavelet Transform and AdaBoost with Random

Forests. Neurocomputing, 177, 188–197.

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

Menggunakan Metode Fuzzy C-Means Clustering

(FCM) Dan Adaptive Neighborhood Modified

Identiﬁcation of Alzheimer’s Disease in MRI Data using Discrete Wavelet Transform and Support Vector Machine

203

Backpropagation (ANMBP). Jurnal Matematika

"MANTIK", 1(1), 31–36.

Pandey, A., Chandra, R., Dutta, M. K., Burget, R., Uher,

V., & Minar, J., 2016. Automatic detection of red

lesions in Diabetic Retinopathy using Shape based

extraction technique in fundus image. Conference:

2016 39th International Conference on

Telecommunications and Signal Processing (TSP),

538–542.

Janani, S., Marisuganya, R., and Nivedha, R., 2013. MRI

Image Segmentation using Stationary Wavelet

Transform and FCM Algorithm Under the guidance

of. International Journal of Computer Applications

(0975–8887), International Conference on Innovations

in Intelligent Instrumentation, Optimization and

Signal Processing “ICIIIOSP-2013”, pp. 1–4.

The National Academy on an Aging Society, 2000.

Alzheimer ’s Disease and Dementia.

Wilson, D., & Laxminarayan, S., 2007. Handbook of

Biomedical Image Analysis: Volume II : Segmentation

Models, Part B. New York: Kluwer Academic/

Plenum Publishers.

Zhang, D., Wang, Y., Zhou, L., Yuan, H., & Shen, D.,

2011. Multimodal Classification of Alzheimer’s

Disease and Mild Cognitive Impairment. Neuroimage,

55(3), 856–867, doi:

10.1016/j.neuroimage.2011.01.008.

ICMIs 2018 - International Conference on Mathematics and Islam

204