Heart Disease Detection using Iridology with Principal Component

Analysis (PCA) and Backpropagation Neural Network

Leonardus Sandy Ade Putra

, R. Rizal Isnanto

, Aris Triwiyatno

and Vincentius Abdi Gunawan

Department of Informatics Engineering, STMIK Palangka Raya, Palangka Raya, Indonesia

Department of Computer Engineering, Diponegoro University, Semarang, Indonesia

Department of Electrical Engineering, Diponegoro University, Semarang, Indonesia

Department of Informatics Engineering, Palangka Raya University, Palangka Raya, Indonesia

Keywords: Iridology, Iris, Heart Disease, Principal Component Analysis and Backpropagation Neural Network.

Abstract: Heart is one of many vital organs on the human body which function is to pump blood throughout the body.

Based on the data from World Health Organization (WHO), impaired heart function is the number one

cause of death in the world. Early symptoms of heart disease commonly go unnoticed by the patients

themselves and are often neglected. According to some circles on society, heart condition checking is

assumed expensive, inconvenient, and takes a long time to do. A simpler and cheaper way to detect early

heart complication symptom is needed. The iridology method can be used as a solution to resolve the

problem above. Iridology is a method to determine abnormalities or complications that are happening on an

organ’s function by taking an image on iris as the main object of diagnosis. This research is done to make a

system using image processing, feature extraction using Principal Component Analysis (PCA) and

classification using Backpropagation Neural Network to recognize the condition of the heart’s function. The

researcher used 90 patient data with normal and abnormal heart condition. These data will be divided into

50 training data and 40 test data. Based on the test that has been done by using PCA score result variations

as many as 600, 500, 400, 300 and 200, percentages of recognition rate have been obtained. The percentages

in order are 92.5%, 90%, 85%, 75%, and 67.5%. The designed system can be used to detect early symptoms

of heart function problem by using the Iridology method with the highest recognition rate of 92.5% using

the PCA score of 600.

1 INTRODUCTION

Heart is an internal organ which function is really

important for the human’s body. The heart is very

vulnerable to have a failure on their function on

distributing blood to the whole body. According to

World Health Organization (WHO), heart

complication or Cardiovascular disease (CVD) is the

number one cause of death in the world. CVD takes

17.7 million lives each year, 31% of all global

deaths (World Health Organization, 2017). From the

10 deadliest diseases in Indonesia, the CVD disease

stands on the first rank for the cause of death (Tv,

2015). Coronary heart sufferers in Indonesia reaches

7.6 million people per year (Kematian, Jantung, &

Tinggi, 2018). The main factors that trigger heart

disease are tobacco usage, unhealthy diet program,

lack of exercise, and alcohol consumption. To some

people, heart condition checking is assumed as

something expensive, complicated, and needs a lot

of time. Therefore, a more convenient and cheaper

way to detect early symptoms of heart complication

is needed.

Iridology is a method to identify abnormalities or

disturbances which are happening on the organs of

the body by making use of colours, structures,

patterns, and fibers which can be seen on the iris

(Samant & Agarwal, 2017). Iridology has shown a

great result regarding the potential of Iridology

method on analyzing human organs (Ernst, 1999;

Samant & Agarwal, 2017; Sitorus, Purnomo, &

Wibawa, 2016; Wibawa & Purnomo, 2006). There

were some studies that made use of the iridology

method as a way to detect abnormalities on human

organs. The observation was done by Ignaz Von

Peczely on the iris of an owl. In the iris of the owl,

there are dark spots which previously did not exist.

After the owl was cured, the dark spots are gone

Putra, L., Isnanto, R., Triwiyatno, A. and Gunawan, V.

Heart Disease Detection using Iridology with Principal Component Analysis (PCA) and Backpropagation Neural Network.

DOI: 10.5220/0009009402570264

In Proceedings of the 7th Engineering International Conference on Education, Concept and Application on Green Technology (EIC 2018), pages 257-264

ISBN: 978-989-758-411-4

257

(Frank, Ferreira, & Pellow, 2013). A study by Nils

Lilijequist was done on a patient with lymph

disorder. After the patient did some treatment, the

patient was observed once more and many changes

were found on the iris of the patient (Frank et al.,

2013). Research was also done to diabetic

retinopathy patients which used GLCM feature

extraction and Support Vector Machine (SVM)

classification method, producing the highest success

rate of 88% (Labhade, Jyoti Dnyaneshwar, L. K.

Chouthmol, 2016). Research on diabetic patients

using statistic method and 2D-DWT feature as

feature extraction and Random Forest (RF) as

classification produced a success rate of 89,66%

(Samant & Agarwal, 2017). Research on heart

disease patients using PCA and SVM produced a

success rate of 80%. (Permatasari, Novianty, &

Purboyo, 2017). Detection of diabetes on pancreas

using GLCM and Back propagation Neural Network

produced a success rate of 81,35% (Adelina, Sigit,

Harsono, & Rochmad, 2017). Research was also

done to detect stomach disorder using PCA and

Back propagation Neural Network, resulting to a

success rate of 87,5% from 40 iris images. (Dewi,

Novianty, & Purboyo, 2017).

This research is done to make a system which

can identify any heart function problems based on

the structure of the picture by utilizing image

processing, feature extraction using Principal

Component Analysis (PCA), and Classification

using the Backpropagation Neural Network.

2 RESEARCH METHOD

2.1 Iridology

Iridology is a method to recognize an organ’s

condition and the body’s system through

characteristics or signs that appear on the iris

(Aisyah & Dewi, 2016) and as an alternative

medical check up to detect disease or problem on a

specific organ through colour observation

(Permatasari et al., 2017). Usually, Iridology is also

known as iris diagnosis which on the medical world

states that each part of the body can be represented

by an area contained in the iris (Adelina et al.,

2017).

Iridology diagram documents the left and right

eye which reflects the systems and organs’ condition

based on the iris zone which was developed by Dr.

Bernard Jensen. Based on Dr. Bernard Jensen’s

chart, the heart is only on the left iris. The heart’s

position on the left iris is shown on zone 02.10 –

03.10 and can be seen on Figure 1 (“Left Eye Iris

Iridology Chart _ Iridology Chart,” n.d.).

Figure 1: Left iris ridology chart.

2.2 Image Processing

Image processing is done to separate the iris from

the pupil and sclera and then normalize the iris to the

standard dimension which can be adjusted with the

iris chart. This process needs extraction feature and

classification mechanism to conclude the correct

diagnosis. To create a process for detecting the

center of the iris and the center of the pupil, colour

images will be changed into grey images which are

segmented and transformed to the polar coordinates

(Nusantara, Herlambang, Isnanto, & Z, 2015).

2.3 Principal Component Analysis

Principal Component Analysis (PCA) (Nasseri,

Shirazi, & Sadeghigol, 2011) states mathematically

as an orthogonal linear transformation that changes

data to new coordinates system, which means that

PCA exchanges theoretically as an optimal linier

sceme. An image which is shown in a form of linier

projection in line with the eigen vector which

corresponds to the order of the eigen images from

the biggest to the smallest on covarian matrix. Each

eigen vector has one eigen point. Reduction decides

whether the eigen vector will be seen by choosing

from the biggest to the smallest. The characteristic

of data which is reduced has a small Eigen (Bishop,

2013; Duda, Hart, & Stork, 2012; Iridology, 2016;

Smith, 2002).

EIC 2018 - The 7th Engineering International Conference (EIC), Engineering International Conference on Education, Concept and

Application on Green Technology

258

2.4 Backpropagation Neural Network

Backropagation Neural Network is one of the many

artificial representations of the human’s brain that

always tries to simulate the learning process of the

human brain (Rochmad, 2006). This algorithm does

two steps of calculation, the sophisticated

calculation can calculate the mistakes between the

actual output to target and back propagation to

spread the mistakes in order to fix the synaptic

weight on every neuron, this consists of many

layers (multiplayer network) (Aisyah & Dewi,

2016), (Saputra, Tulus, Zarlis, Sembiring, &

Hartama, 2017) can be seen on Figure 2:

1. Input layer (1 piece), consists of 1 X input

unit.

2. Hidden layer (at least 1 piece), consists of 1

hidden Y unit.

3. Output layer (1 piece), consists of 1to M

output unit.

Figure 2: Backpropagation neural network architecture.

3 RESULT AND DISCUSSION

This research used various PCA scores to identify

the effects on the heart problem recognition stage by

using iris image. The stages of this research can be

seen on Figure 3. The first step is to take a picture of

the left iris, this 1280 x 800 pixel sized picture will

go into the preprocessing process. The preprocessing

process is done to fix the quality of the image and to

seperate the part of the image which is necessary

from the part that is not. After the preprocessing

process, feature extraction will be done to the image

by using PCA. Feature extraction is needed to

simplify a data by maintaining important data

values. Feature extraction with PCA is done with

PCA score variations as many as 600, 500, 400, 300,

and 200.

The result of PCA feature extraction will be

classified using Backpropagation Neural Network in

accordance with the PCA score variation on each

images. This classification is done to group every

pixel on an image so that it can be interprated as a

specific property. The classification of the result is

in a form of information regarding the condition of

normal and abnormal heart condition.

Figure 3: Diagram of flow detection of heart problems.

3.1 Data Collection

This research used 90 datas of the left iris. The datas

are divided into two parts, 50 data as training data

and 40 data as test data.

3.2 Preprocessing

Preprocessing is needed to fix the quality of the

image and to seperate the part of the image which is

necessary from the part that is not. In preprocessing,

the image will go through a few stages before going

through the feature extraction process. The first

process on preprocessing is converting RGB type

image to grayscale type image. Afterwards, the

localization process, a process to determine the

needed location or part of the image, will be done.

The next process is the normalization process.

This process is done to change the shape of the

image from polar shape to square shape with the size

of 81 x 31 pixels. The quality of the image will be

fixed on the contrast enhancement process by using

Heart Disease Detection using Iridology with Principal Component Analysis (PCA) and Backpropagation Neural Network

259

CLAHE. After the images went through every

preprocessing stages, the images are ready for

feature extraction. Figure 4 shows the flowchart of

each preprocessing stages.

Figure 4: Preprocessing stage.

3.2.1 Region of Interest (ROI)

In recognizing iris, ROI is known as a region filled

with complete information of an iris (Li, Li, & Ma,

2012). This process is done to find a part that can be

examined by separating the part (Adelina et al.,

2017). ROI is used as the iridology map (Prayitno,

Wibawa, & Purnomo, 2017), where the heart’s

location is on the left iris with the direction at 02.10

– 03.10 and can be seen on Figure 5. The lines show

ROI in the heart.

Figure 5: ROI from heart organs.

3.2.2 Image Conversion RGB to Grayscale

The first image is a RGB image, which is why a

conversion from the RGB image to grayscale image

is needed to be able to get processed on the next

step. The yellow line indicates the heart’s part on the

iris. The result of the image coversion is displayed

on Figure 6.

Figure 6: Conversion RGB to Grayscale.

3.2.3 Localization

On this step, the function of localization is to

seperate the iris from the eye image. This is done so

that the iris can be processed on the next step,

Normalization. The result of localization can be seen

on Figure 7.

Figure 7: Result of localization.

3.2.4 Normalization

Normalization is a process that changes the shape of

iris, from the polar shape to a square shape. The

square shape will make the area of the iris to the

same size. Which is why some images from the iris

with different sizes will have the same size and also

have the same characteristics with the same location

(Adelina et al., 2017). The center of the pupil is

considered as the reference point and vector radial

that passes through the iris area, which is illustrated

on Figure 8 (Jogi & Sharma, 2014).

Figure 8: Result of normalization.

EIC 2018 - The 7th Engineering International Conference (EIC), Engineering International Conference on Education, Concept and

Application on Green Technology

260

3.2.5 Contrast Enhancement

To improve the contrast of the iris image, the

histogram equalization step is needed. In the

histogram of a normalized iris, the gray level is

concentrated on the center of the gray level from 0

to 255. Contrast Limited Adaptive Histogram

(CLAHE) is used as contrast and threshold

equalization to make the next progress easier (Jogi

& Sharma, 2014). CLAHE is needed to resolve low

contrast and different lightning problems. CLAHE

can decrease or get rid of the noises that are on the

image. The result of the Histogram Equalization

using CLAHE can be seen on Figure 9.

Figure 9: Histogram equalization result using CLAHE.

3.3 Principal Component Analysis

To decrease the dimension of the database, the

function of PCA is to mantain the characteristics of

the dataset, (Liu & Wechsler, 2000) which are made

of high variations. Figure 10 shows an image of the

preprocessing result with the size of 81 x 31 pixel

which PCA value is ready to be found. Steps on how

to apply PCA can be seen on Table 1. Tables must

appear inside the designated margins or they may

span the two columns.

Figure 10: The image of the preprocessing result.

Table 1: Steps for Implementing PCA.

Steps

Explanation

Taking input

pictures (X)

Image of the preprocessing result

(81 x 31) is transposed to matrix

with each image changed to 1 x

2511.

50 data are available so the size

of the combined matrix will be

(50 x 2511).

Counting Mean 𝑿

Counting the average matrix (X)

on each column.

Counting zero

mean (Z)

Zero mean = (X) - (𝑿

)

Counting

covariance (C)

C = 𝒁

𝑻

∗ 𝒁

Counting eigen

vectors and eigen

value

[ v , d ] = eig (C)

eig = (Z*v)

The size of eigen value is 1 x

2511 and the size of eigen vector

is 50 x 2511.

PCA

Eigen value has been sorted

from the biggest to the smallest

and then the eigen vector will be

searched according to the score

order of the eigen value.

3.4 Principal Component Analysis

Backpropagation neural networks is used to find the

best result from the classification process. Data

image is divided into 2, for training and test. 50 data

which contain 25 normal data and 25 abnormal data

will be used as training data. This research used 2

hidden layers. The parameter which was set for this

research is on Table 2.

Table 2: Backpropagation neural network.

Parameters

Number of neurons

[10 15]

Maximum Epoch

1000

Targets

1e-6

Learning rate

0.0001

3.5 Data Testing

On the testing step, images will be classified and

produce information about the heart’s condition.

Condition in which the classifcation result is under 1

is considered abnormal or has a heart problem, while

the classification result which is above 1 is

considered normal or does not have any heart

problem. The test which was done with many score

variations can be seen on Figure 11.

Heart Disease Detection using Iridology with Principal Component Analysis (PCA) and Backpropagation Neural Network

261

Figure 11: Detection rate of heart disease.

The usage of PCA score variations as many as

600, 500, 400, 300, and 200 is involved in this

research. The PCA score is the most important score

from the feature extraction of the original image and

as less value. The less PCA value has already

represented a valuable information from the original

image. The diagnosis level of the condition of the

heart in order are 92.5%, 90%, 85%, 75%, and

67.5%. It can be seen that the PCA score variation

affects the heart condition diagnosis system success

rate. The more PCA score that is used, the higher the

chance of success.

As an example to find out the result of the test on

this system, researcher will show the result of the

classification by using Backpropagation Neural

Network on the test system using the PCA score of

600 which can be seen on Table 3.

Table 3: Test result with the PCA score of 600.

Sample

Test Value

Expected

Test

Output

Result

1.JPG

0.403148

Abnormal

2.JPG

0.129456

Abnormal

3.JPG

0.083802

Abnormal

4.JPG

0.058263

Abnormal

5.JPG

0.66082

Abnormal

6.JPG

0.44757

Abnormal

7.JPG

0.662363

Abnormal

8.JPG

0.707559

Abnormal

9.JPG

0.837575

Abnormal

10.JPG

0.899931

Abnormal

11.JPG

0.0090691

Abnormal

12.JPG

0.816724

Abnormal

14.JPG

0.491388

Abnormal

15.JPG

0.656056

Abnormal

16.JPG

0.409854

Abnormal

17.JPG

2.03304

Abnormal

Normal

18.JPG

0.953222

Abnormal

Normal

19.JPG

0.513437

Abnormal

20.JPG

0.523197

Abnormal

21.JPG

4.45045

Normal

22.JPG

1.62642

Normal

23.JPG

3.49232

Normal

24.JPG

1.88061

Normal

25.JPG

3.22679

Normal

26.JPG

1.46215

Normal

27.JPG

3.53182

Normal

28.JPG

1.02171

Normal

29.JPG

1.51179

Normal

30.JPG

1.21271

Normal

31.JPG

1.20584

Normal

32.JPG

3.37536

Normal

33.JPG

1.08617

Normal

34.JPG

1.86534

Normal

35.JPG

1.48094

Normal

36.JPG

1.31759

Normal

37.JPG

1.13088

Normal

38.JPG

0.1638

Normal

Abnormal

39.JPG

1.52222

Normal

40.JPG

1.21769

Normal

From the results above shows that the main

component analysis method as feature extraction and

backpropagation neural network as classification has

better results than the method used by (Permatasari

et al., 2017) in diagnosing heart conditions through

iris.

4 CONCLUSIONS

Based on the test result of PCA score variation, it

can be seen that the score of PCA has affected the

recognition rate of early heart problem symptoms.

The more the score of PCA is, the higher the

successful rate is and it is inversely propotional with

the usage of less score.

By using image processing, feature extraction by

using PCA, and classification by using

Backpropagation Neural Network, the designed

system could work well. Tests on a few PCA score

variations as many as 600, 500, 400, 300, dan 200

has been tested on 40 test data and produced a

succesful rate of 92.5%, 90%, 85%, 75%, dan

67.5%.

Therefore the designed system can recognize

early heart function problem using the Iridology

method with 92.5% as the highest rate of success by

using the PCA score of 600.

REFERENCES

Adelina, D. C., Sigit, R., Harsono, T., & Rochmad, M.

(2017). Identification Of Diabetes In Pancreatic

Organs Using Iridology, International Electronics

Symposium on Knowledge Creation and Intelligent

Computing, Indonesia: IES-KCIC, pp 14–119.

EIC 2018 - The 7th Engineering International Conference (EIC), Engineering International Conference on Education, Concept and

Application on Green Technology

262

Aisyah, M., & Dewi, K. (2016). Stomach Disorder

Detection Through the Iris Image Using

Backpropagation Neural, (Icic), pp 1–6.

Bishop, C. M. (2013). Pattern Recognition and Machine

Learning. Journal of Chemical Information and

Modeling, Vol. 53, 049901

Dewi, A. K., Novianty, A., & Purboyo, T. W. (2017).

Stomach disorder detection through the Iris Image

using Backpropagation Neural Network. 2016

International Conference on Informatics and

Computing, ICIC 2016, Indonesia: Icic, pp 192–197.

Duda, R., Hart, P., & Stork, D. (2012). Patterns

Classification. John Wiley & Sons,.

Ernst, E. (1999). Iridology : A Systematic Review,

US National Library of Medicine National Institutes of

Health Search database, Vol 6(1), pp 7–9.

Frank, L., Ferreira, J. T., & Pellow, J. (2013). The validity

and reliability of iridology in the diagnosis of previous

acute appendicitis as evi- denced by appendectomy.

The South African Optometrist, Vol 72(3),pp 127–132.

Jogi, S. P. S. P., & Sharma, B. B. B. B. (2014).

Methodology of iris image analysis for clinical

diagnosis. 2014 International Conference on Medical

Imaging, m-Health and Emerging Communication

Systems, India: MedCom, pp 235–240.

Kematian, A., Jantung, P., & Tinggi, M. (2018). Angka

Kematian Penderita Jantung Masih Tinggi.

Labhade, Jyoti Dnyaneshwar, L. K. Chouthmol, S. D.

(2016). Diabetic Retinopathy Detection Using Soft

Computing Techniques. International Conference on

Automatic Control and Dynamic Optimization

Techniques (ICACDOT), pp 175–178.

Left Eye Iris Iridology Chart _ Iridology Chart. (n.d.).

Li, Y., Li, W., & Ma, Y. (2012). Accurate iris location

based on region of interest. Proceedings - 2012

International Conference on Biomedical Engineering

and Biotechnology, ICBEB 2012, pp 704–707.

Liu, C., & Wechsler, H. (2000). Evolutionary pursuit and

its application to face recognition. IEEE Transactions

on Pattern Analysis and Machine Intelligence, Vol

22(6), pp 570–582.

Nasseri, L., Shirazi, A. A. B., & Sadeghigol, N. (2011).

Tsallis entropy, PCA and neural network in novel

algorithm of iris classification. Proceedings of the 2011

World Congress on Information and Communication

Technologies, WICT 2011, pp 385–390.

Nusantara, R. G. A., Herlambang, P., Isnanto, R. R., & Z,

A. A. (2015). Application of Liver Disease Detection

Using Iridology with Back-Propagation Neural

Network, pp 123–127.

Permatasari, L. I., Novianty, A., & Purboyo, T. W. (2017).

Heart disorder detection based on computerized

iridology using support vector machine. ICCEREC

2016 - International Conference on Control,

Electronics, Renewable Energy, and Communications

2016, Conference Proceedings, pp 157–161.

Prayitno, A., Wibawa, A. D., & Purnomo, M. H. (2017).

Early detection study of Kidney Organ Complication

caused by Diabetes Mellitus using iris image color

constancy. Proceedings of 2016 International

Conference on Information and Communication

Technology and Systems, ICTS 2016, pp 146–149.

Rochmad, M. dkk. (2006). Osteoporosis Symptoms

Detection Through the Iris Using Image Clustering.

EEPIS, (0852-2863).

Samant, P., & Agarwal, R. (2017). Diagnosis of Diabetes

Using Computer Methods: Soft Computing Methods

for Diabetes Detection Using Iris, Vol 200(2), pp 57–

62.

Saputra, W., Tulus, T., Zarlis, M., Sembiring, R. W., &

Hartama, D. (2017). Analysis Resilient Algorithm on

Artificial Neural Network Backpropagation. Journal

of Physics: Conference Series, Vol 930(1).

Sitorus, M. A. R., Purnomo, M. H., & Wibawa, A. D.

(2016). Iris image analysis of patient Chronic Renal

Failure (CRF) using watershed algorithm. Proceedings

- 2015 4th International Conference on

Instrumentation, Communications, Information

Technology and Biomedical Engineering, ICICI-BME

2015, (October 2016), pp 54–58.

https://doi.org/10.1109/ICICI-BME.2015.7401334

Smith, L. I. (2002). A tutorial on Principal Components

Analysis Introduction. Statistics, pp 51- 52.

Tv, C. N. N. (2015). Home Nasional Internasional

Ekonomi Olahraga Teknologi Hiburan Gaya Hidup.

Wibawa, A. D., & Purnomo, M. H. (2006). Early detection

on the condition of Pancreas organ as the cause of

diabetes mellitus by real time iris image processing.

IEEE Asia-Pacific Conference on Circuits and

Systems, Proceedings, APCCAS, pp 1008–1010.

World Health Organization. (2017). Cardiovascular

Disease: World Heart Day 2017. Who.

Heart Disease Detection using Iridology with Principal Component Analysis (PCA) and Backpropagation Neural Network

263