1. Automated Detection of Glaucoma using
K Means clustering algorithm
Project Report
Submitted in partial fulfillment of the requirements
For the degree of
Bachelor of Engineering
(Information Technology)
Name Roll No
Asmita Sarkar 2011-3027
Debangana Dutta 2011-3023
Snehangshu Dan 2011-3052
Debpratim Chakraborty 2011-3022
B.E 4th Year Dept:IT Batch:2011-2015
Under the Guidance of Mr.Arindam Chowdhury
UNIVERSITY INSTITUTE OF TECHNOLOGY
THE UNIVERSITY OF BURDWAN
BURDWAN

2. ACKNOWLEDGEMENT
A project is the result of hours of painstaking work and this is no
exception. There are people who deserve special mention here; without
whom this project would not have been possible.
First and foremost, we would like to thank Asst Prof. Arindam
Chowdhury, for his advice, supervision and crucial contribution, which
have made him the backbone of this project. We would also like to
thank our Head of the Department, Asst Prof. SOUBHIK
BHATTACHARYA for his invaluable advice; and the entire faculty of
the CSE-IT department of University Institute of Technology,Burdwan
for their cooperation and valuable input.
Finally, during a project, there are times when we lose the will to go on,
and wish the project were over then and there. Our friends helped us
endure those times with their unfailing humor and unflinching
encouragements. If we owe more to someone than others, they would
definitely be my parents, for their constant support and blessings,
without whom I would not be where I am today.

3. Certificate from the Project Guide
This is to certify that this project entitled “Automated Detection of Glaucoma
using K Means clustering algorithm” submitted in partial fulfillment of the
degree of B.E.,UNIVERSITY INSTITUTE OF
TECHNOLOGY,BURDWAN UNIVERSITY is done by-
Name Roll No
Asmita Sarkar 2011-3027
Debangana Dutta 2011-3023
Snehangshu Dan 2011-3052
Debpratim Chakraborty 2011-3022
-is an authentic work carried out by them under my
guidance. The matter embodied in this project work has not been submitted
earlier for award of any degree to the best of my knowledge and belief, and the
project report has been developed according to the “B.E. PROJECT &
PROJECT REPORT STANDARD, THE UNIVERSITY OF BURDWAN”.
Signature of the Student Signature of the Guide

4. Self-Certificate
This is to certify that the dissertation/project report entited “Automated
detection of glaucoma using K Means Clustering algorithm”,
done by us, is an authentic work carried out for the partial fulfillment of the
requirements for the award of the degree of B.E, under the guidance of
Asst. Prof Mr.Arindam Chowdhury.
We also certify that We are aware of the “B.E PROJECT & PROJECT
REPORT STANDARD, THE UNIVERSITY OF BURDWAN” issued by the
University of Burdwan, and this project report is based on that standard. The
matter embodied in this project work has not been submitted earlier for the
award of any degree or diploma to the best of our knowledge and belief.
Signature of the
Students
Asmita Sarkar
Debangana Dutta
Snehangshu Dan
Debpratim Chakraborty

5. INDEX
Sl
No
Topics Page
1)
2)
3)
4)
5)
6)
7)
8)
9)
10)
Objective
Introduction
a)What is glaucoma?
b)Risk factors of glaucoma
c)How can glaucoma be detected?
d)What is optic disc & optic cup?
e)What is K Means clustering?
Methodology used in this project
a)Flowchart of Methodology used.
b)RGB Colour extraction
c)G Plane extraction
d)Region of Interest(ROI) extraction
e)Cluster Identification using K Means algorithm
f)Ellipse Fitting
g)Cup to Disc Ratio Calculation(CDR)
Examples of testing of the proposed
methodology on a known normal and
glaucomateous image
a)Normal Image
b)Glaucomateous Image
Results
Comparitive Study
Methods Approached Previously
Conclusion
Future Scope
References
1
2
3-4
5-6
7
8-9
11
12
12
13
13-15
15
16
17-18
19-20
21
22-28
29
30
31
32

7. 1
Objective
To detect glaucoma in an optic image by
calculating the optic cup to optic disc ratio by
segmenting the optic image using k means
clustering.

8. 2
Introduction
What is Glaucoma?
Glaucoma is a complicated disease in which
damage to the optic nerve leads to progressive,
irreversible vision loss. Glaucoma is the second
leading cause of blindness. Glaucoma has been
called the "silent thief of sight" because the loss
of vision often occurs gradually over a long period
of time, and symptoms only occur when the
disease is quite advanced. Once lost, vision cannot
normally be recovered, so treatment is aimed at
preventing further loss. Worldwide, glaucoma is
the second-leading cause of blindness after
cataracts. It is also the leading cause of blindness
among African Americans. Glaucoma affects one
in 200 people aged 50 and younger, and one in 10
over the age of 80. If the condition is detected
early enough, it is possible to arrest the
development or slow the progression with
medical and surgical means.

9. 3
Risk Factors of Glaucoma:
Because chronic forms of glaucoma can
destroy vision before any signs or symptoms
are apparent, be aware of these factors:
 Elevated internal eye pressure
(intraocular pressure).- If your internal eye
pressure (intraocular pressure) is higher than
normal, you're at increased risk of developing
glaucoma, though not everyone with elevated
intraocular pressure develops the disease.
 Age- You're at a higher risk of glaucoma if
you're older than age 60, particularly if you're
Mexican-American. You may be at higher risk
of angle-closure glaucoma if you're older than
age 40

10. 4
 Medical conditions- Several conditions
may increase your risk of developing
glaucoma, including diabetes, heart diseases,
high blood pressure and hypothyroidism.
 Other eye conditions-Severe eye injuries
can cause increased eye pressure. Other eye
conditions that could cause increased risk of
glaucoma include eye tumours, retinal
detachment, eye inflammation and lens
dislocation. Certain types of eye surgery also
may trigger glaucoma. Also, being near
sighted or farsighted may increase your risk of
developing glaucoma.
 Long-term corticosteroid use-Using
corticosteroid medications, especially eye
drops for a long period of time may increase
your risk of developing secondary glaucoma

11. 5
How can glaucoma be detected?
1.Elevated Intraocular pressure (>22 mmHg),
2. Increased cup-to-disc ratio(The cup-to-disc ratio
compares the diameter of the "cup" portion of the
optic disc with the total diameter of the optic disc.
A good analogy to better understand the cup-to-
disc ratio is the ratio of a donut hole to a donut.
The hole represents the cup and the surrounding
area the disc. If the cup fills 1/10 of the disc, the
ratio will be 0.1. If it fills 7/10 of the disc, the ratio
is 0.7. The normal cup-to-disc ratio is 0.3. A large
cup-to-disc ratio may imply glaucoma or other
pathology), and
3.The ISNT rule – It states that in a healthy optic
disc, the widest rim tissue is found inferiorly, then
superiorly, nasally, with the temporal rim being
the thinnest. This gives rise to a cup shape that is
often a horizontal oval.

12. 6
We will use cup to disc diameter ratio for
detecting the possibility of glaucoma in the image
of an eye.
If the cup to disc ratio is greater than 0.3,the
patient is doubted to be suffering from glaucoma
and may be referred for further treatments.

13. 7
What is optic disc and cup?
The optic disc is the anatomical location of the
eye's “blind spot", the area where the optic nerve
and blood vessels enter the retina . The optic disc
can be flat or it can have a certain amount of
normal cupping. Glaucoma produces additional
pathological cupping of the optic disc.
Fig-illustration of optic disc and
optic cup in an optic image

14. 8
What is K Means Clustering?
K-means (MacQueen, 1967) is one of the simplest
unsupervised learning algorithms that solve the
well known clustering problem. The procedure
follows a simple and easy way to classify a given
data set through a certain number of clusters
(assume k clusters) fixed a priori. The main idea is
to define k centroids, one for each cluster. These
centroids shoud be placed in a cunning way
because of different location causes different
result. So, the better choice is to place them as
much as possible far away from each other. The
next step is to take each point belonging to a
given data set and associate it to the nearest
centroid. When no point is pending, the first step
is completed and an early groupage is done. At
this point we need to re-calculate k new centroids
as barycenters of the clusters resulting from the
previous step. After we have these k new
centroids, a new binding has to be done between
the same data set points and the nearest new
centroid. A loop has been generated. As a result

15. 9
of this loop we may notice that the k centroids
change their location step by step until no more
changes are done. In other words centroids do
not move any more.
Finally, this algorithm aims at minimizing an
objective function, in this case a squared error
function. The objective function
,where is a chosen distance measure
between a data point xi
(j)
and the cluster centre cj,
is an indicator of the distance of the n data points
from their respective cluster centres.

16. 10
Methodology used in this project-
This work uses K means clustering algorithm to
segment optic disc and optic cup from an optic
image and measure the optic cup and optic disc
ratios.
The work consists of mainly three stages of
operation as given below-
1.G plane extraction from the optic image
2.ROI(Region of Interest) extraction
3.Measurement of optic cup to optic disc ratio by
calculating their respective areas using ellipse
fitting.
All the stages are explained below in details.

17. 11
Flowchart of the methodology used-
This work is based on the following flowchart
The important steps used in the flowchart are
explained in the next pages

18. 12
1.RGB Colour Extraction
The first step is to take the input image and
separate the colour components of the grayscale
image into it’s red,blue and green components.
It is necessary for extracting the G component
only which helps in further processing of the
image for ROI extraction.
2.G Plane extraction
G plane is considered for the extraction of optic
disc and optic cup, because G plane provides
better contrast than the other two planes Red and
Blue planes. Hence it is necessary to separate the
G plane for further analysis.

19. 13
3. Region of Interest (ROI Extraction)
After the extraction of the G Plane, we have to
take the region of interest for this project (which
is the optic disc and the optic cup).Thus, we have
to eliminate the outer regions of the eye and take
only the desired portion to be used for optic cup
to disc ratio.
4. Cluster identification using K Means algorithm
K-means clustering plays a vital role in the feature
extraction stage to compute the features (CDR). It
is an unsupervised learning algorithm that solves
the well-known clustering problem. The
procedure follows a simple and easy way to
classify a given data set through a certain number
of clusters (k clusters) fixed a priori.
This work implements a k-means algorithm that
finds the centres of n clusters and groups the
input samples around theclustersto define k
centroids, one for each cluster. At this point k new
centroids are calculated as the mean of

20. 14
the clusters resulting from the previous step.
As a result of repetitive application of these
two steps, the k centroids change their
location step by step until no more changes
take place.
The K-means algorithm is an iterative
technique that is used to partition the ROI
image into K clusters. The ROI covers mainly
the entire optic disc, optic cup and a small
portion of other regions of the image. Hence
the K value is chosen as 3.
From the three clusters as shown in the optic
disc cluster has to be identified as follows. The
cluster which contains the border region
belongs to the other region of the ROI image
that does not contain the optic disc and optic
cup. Hence, the cluster is removed for the
extraction of optic disc and the remaining two
clusters form the optic disk region. Since it is
clearlyknown that the optic cup is inside the
optic disc,

21. 15
The cluster in the center of the image forms the
optic cup .
After extracting optic disc, contours are found and
the contour with maximum area will correspond
to the optic disc.
5. Ellipse fitting
Rectangles are drawn for each optic disc and cup
areas enclosing their regions.then an ellipse is
drawn for each of the cup and disc areas
,inscribed within the rectangle.
The area of the ellipse as well as the area of the
optic disc is calculated by using the formula A =
πab where a and b are major axis length (half of
the rectangle width) and minor axis length (half of
the rectangle height), respectively. The area of
optic cup is also computed in the same manner.

22. 16
6.Cup to Disc ratio calculation (CDR).
The CDR which is the ratio between the area
of the optic cup and the area of the optic disc
is computed and used as the feature for the
detection of glaucoma.
Cup to disc ratio=Area of optic cup/Area of optic disc
If the cup to disc ratio comes above 0.3, the
patient can be suspected to have glaucoma. If
the ratio comes below 0.3, the patient maybe
considered safe.

23. 17
Examples of Testing of the proposed
methodology on a normal and a glaucoma image
(known)
1.Normal image
A normal image is taken from a standard
dataset and testing is done as follows.
Normal eye
image
G plane extraction

24. 18
Region of
interest
K
clustered
Area Calculation by ellipse
fitting
Next, the CDR is calculated, which comes to
be 0.27 for this image thus confirming with
the dataset that it is a normal, non
glaucomateous image.

25. 19
2.Glaucomateous image
A suspected glaucoma image is taken from a
standard database and the steps of the project
are performed as follows-
Glaucomateous image
G plane of the glaucomateous image

26. 20
ROI Extraction of the glaucomateous
image
K means clustered glaucoma image
Area by curve fitting of the
glaucomateous image
Next,the CDR is calculated which comes to be 0.4,
thus confirming with the dataset that it may be
glaucomateous.

27. 21
Results-
The proposed system is tested upon a batch of
455 images taken from a standard dataset of
normal and glaucomateous images.
Regions were correctly identified in 374 images.
The region detection failure rate is 18%.
Accuracy of the system is 60%(out of 374 images)
Accuracy
Total
455
81 images.
225 correct
detection
149
incorrect
detection
could not
detect

28. 22
Comparitive Study-
Two previous papers on optic disc and cup
segmentation using k means clustering have been
consulted for this project-
1.An efficient automated sysyem for glaucoma
detection using fundus image
By K Narasimhan, Dr. K Vijayrekha
2. Clustering Based Optic Disc and Optic Cup
Segmentation for Glaucoma Detection
V.Mahalakshmi ,S.Karthikeyan

29. 23
The flowcharts of the above two papers
respectively are-
Flowchart of first paper
Flowchart of second paper

30. 24
As can be noticed, there is not much difference in
the methods of their process and this project,but
some of the techniques used were different in the
three methods. Also, the accuracy of the three
methods are different due to the different
methods used.Thecomparisioncan be explained
as follows-
With Respect to First paper by By
K.NARASIMHAN, Dr.K.VIJAYAREKHA-
1.For the k means clustering their paper uses
three classifiers for the clusterization- KNN, SVM
and Bayes classifier are used to analyze the
performance of the proposed system.
Our System uses only BAYES Classifier for this. In
Bayes classifier, normal Gaussian distribution is
used to fit or model the feature base and it
estimates the prior probabilities from the relative
frequencies of the classes in training.

31. 25
2. They also added features of ISNT Ratio in their
methodology
Results-
We have compiled our project on a standard
database for testing our proposed system which
consisted of 455 images.
The accuracy rate achieved in our project is nearly
60%
They have used A batch of 36 retinal images
obtained from the Aravind Eye Hospital,
Madurai, Tamilnadu, India to assess the
performance of the proposed system.
Using BAYES Classification they obtained 66%
accuracy,

32. 26
With Respect to Second Paper byV.Mahalakshmi ,
S.Karthikeyan
The Main points which can be compared are-
1,They used Gabor Filter to reduce noise and edge
detection.It was necessary because they took
their dataset using a digital camera which may
have inherited a lot of noise . But our process has
a dataset which was very low in noise. Using
Gabor Filter may have increased complexity.
2.They have used vertical diameters for
calculating cup to disc ratios whereas we have
used ellipse fitting for area calculation and
calculated the ratio of the areas of optic disc and
cup.
3. They didn’t use G plane extraction which we
used, being a project in java
Results-
They have used only 8 normal and 8 abnormal
images.

33. 27
Images are collected from Mahatma Eye care
Hospital,Trichirapalli . The images were acquired
using a Canon CR5 non–mydriatic 3 CCD camera
with a 45 degree field of view(FOV).Each image
was captured using 8 bits per colour plane at 768
by 584 pixels.The FOV of image is circular with a
diameter of approximately 540 pixels. For this
database, the images have been cropped around
FOV.For each image, a mask image is provided
that delineates the FOV.
They had nearly 61% accuracy using their method.
We have used nearly 455 images and used the
standard database from the net.
We have an accuracy of nearly 60 %

34. 28
Conclusion from the comparision-
This work confirms nearly to the accuracies of the
previous works done with this K Means
clustereing method and thus can be considered
safe for testing optic images for the detection of
glaucoma.

35. 29
Methods approached previously
Before K Means clustering, level set segmentation
was approached to calculate CDR.
But, level set segmentation was very complex to
implement as it is hugely based on partial
differential equations which could not be
programmed well enough.
K Means algorithm is easier to code and huge
number of medical images can be tested in this
algorithm in a short span of time reducing time
complexity which was harder in level set
segmentation.

36. 30
CONCLUSION-
This paper presents K Means Clustering based
optic disc and cup segmentations for glaucoma
detection. Structural features like CDR (Cup to
Disc Ratio) are measured and if the ratio value
exceeds 0.3 shall be recommended for further
analysis of a patient This shall help in patients
worldwide by protecting further vision
deterioration through timely medical
intervention. We can increase the number of
patients and analyze the performance.

37. 31
Future Scope-
1.ISNT Method can be used additionally with
this work to increase accuracy of the work.
2.K medians algorithm can be approached but
the boundaries in k medians approach may
not be clear
3.Many other modifications of this project can
be made based on situation.

38. 32
References-
1.Digital Image Processing by Gonzalez
2.Wikipedia
3. An efficient automated system for glaucoma
detection using fundus image
K. Narasimhan,Dr. K Vijayarekha
Asstt Prof., Department of ECE, SASTRA
Assoc. Dean., Department of EEE, SASTRA
4. Clustering Based Optic Disc and Optic Cup
Segmentation for Glaucoma Detection
V.Mahalakshmi ,S.Karthikeyan
PG Student, Department of Electronics and
Communication Engineering, K. S. R. College of
Engineering, Tiruchengode, IndiaProfessor,
Department of Electronics and Communication
Engineering, K. S. R. College of Engineering,
Tiruchengode, India

39. 33

40. 34