IRJET - Classification of Retinopathy using Machine Learning

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 04 | Apr 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 1968
Classification of Retinopathy using Machine Learning
Swetha S1, Abarna A2, Sornadevi G3, Kamalesh S4
[1]-[3]Student, Department of IT, Velammal College of Engineering and Technology, Tamilnadu, India.
[4]Professor, Department of IT, Velammal College of Engineering and Technology, Tamilnadu, India.
------------------------------------------------------------------------***-----------------------------------------------------------------------
Abstract -Retinopathy is the most common cause of
blindness of the eye depends on hypertension, diabetes,
prematurity of a person. For this reason, early detection of
retinopathy is of critical importance. In this study, the digital
image processing and a machine learning-based approach
plays an important role in the medical field for detection of
various symptoms of diseases for the early detection of
retinopathy from retinal images. Here we are using feature
extraction and machine learning algorithm “K-Nearest
Neighbor” used for predicting and classifying disease on
retinal images. The accuracy is based on the algorithm which
we are going to use and the datasets and splitting them into
train set and test set.
Key Words: Diabetic retinopathy, Hypertensive
retinopathy, Retinopathy of prematurity, Healthy
retinopathy Image processing, Convolution neural
network.
1. INTRODUCTION
Retinopathy is any damage to the retina of the eyes, which
may cause vision impairment. Retinopathy often refers to
retinal vascular disease, or damage to the retina caused by
abnormal blood flow. Retinopathy occurs when blood
vessels in the back of the eye, the retina, become damaged.
When the blood vessels become damaged they can leak
and these leaks can cause dark spots on our vision. The
main causes of retinopathy tend to be sustained high blood
glucose levels and high blood pressure.
Hypertension is probably the best known systemic
condition associated with non-diabetic retinopathy. In
people with hypertension, retinopathy is often referred to
as hypertensive retinopathy, although this definition has
sometimes been expanded to include retinal arteriolar
signs such as arteriovenous nicking and focal and
generalized arteriolar narrowing. Retinopathy has been
found to be present in about 11% of hypertensive non-
diabetic people over 43 years of age. An appreciable
proportion (average 6%) of normotensive non- diabetic
people may also have retinopathy.1-3 In two population
based studies, more than 50% of the participants with
non- diabetic retinopathy did not have a history of
hypertension.[2],[3] Retinopathy may thus represent the
cumulative effects of elevated blood pressure throughout
life in people not classified as having hypertension.
Diabetic retinopathy also known as diabetic eye disease
is a medical condition in which damage occurs to the
retina due to diabetes mellitus. It is a leading cause of
blindness in developed countries.
[1]Diabetic retinopathy affects up to 80 percent of those
who have had diabetes for 20 years or more.[2] At least
90% of new cases could be reduced with proper
treatment and monitoring of the eyes.[3] The longer a
person has diabetes, the higher chances of developing
diabetic retinopathy.[4] Each year in the United States,
diabetic retinopathy accounts for 12% of all new cases of
blindness. It is also the leading cause of blindness in
people aged 20 to64.
The first stage, called non-proliferative diabetic
retinopathy (NPDR), has no symptoms. Patients may not
notice the signs and have 20/20 vision. The only way to
detect NPDR is by fundus photography, in which
microaneurysms(microscopic blood-filled bulges in the
artery walls) can be seen. If there is reduced vision,
fluorescein angiography can show narrowing or blocked
retinal blood vessels clearly (lack of blood flow or retinal
ischemia).
Macular edema, in which blood vessels leak their contents
into the macular region, can occur at any stage of NPDR.
Its symptoms are blurred vision and darkened or
distorted images that are not the same in both eyes. Ten
percent (10%) of diabetic patients will have vision loss
related to macular edema. Optical Coherence Tomography
can show areas of retinal thickening due to fluid
accumulation from macular edema.
[5]In the second stage, abnormal new blood vessels
(neovascularization) form at the back of the eye as part of
proliferative diabetic retinopathy (PDR); these can burst
and bleed (vitreous hemorrhage) and blur the vision,
because these new blood vessels are fragile. The first time
this bleeding occurs, it may not be very severe. In most
cases, it will leave just a few specks of blood, or spots
floating in a person's visual field, though the spots often
go away after a few hours.
These spots are often followed within a few days or weeks
by a much greater leakage of blood, which blurs the
vision. In extreme cases, a person may only be able to tell
light from dark in that eye. It may take the blood anywhere

from a few days to months or even years to clear from the
inside of the eye, and in some cases the blood will not clear.
Retinopathy of prematurity (ROP), also called
retrolental fibroplasia (RLF) and Terry syndrome, is a
disease of the eye affecting prematurely born babies
generally having received intensive neonatal care, in which
oxygen therapy is used on them due to the premature
development of their lungs. It is thought to be caused by
disorganized growth of retinal blood vessels which may
result in scarring and retinal detachment. ROP can be mild
and may resolve spontaneously, but it may lead to
blindness in serious cases. Thus, all preterm babies are at
risk for ROP, and very low birth-weight is an additional
risk factor. Both oxygen toxicity and relative hypoxia can
contribute to the development of ROP.
Healthy corresponds to people who do not suffer from any
of the above retinopathy types.
a)Hypertensive b)Diabetic C)Premature D)Healthy
Fig-1: Retinopathy fundus images
2. LITERATURE SURVEY
2.1Retinal vascular development in premature
infants
Retinopathy of prematurity (ROP) is a proliferative retinal
vascular disease affecting the premature infant with an
incompletely vascularized retina. The spectrum of
ophthalmological findings in ROP exists from minimal
sequel, which do not affect vision, to bilateral retinal
detachment and total blindness. With the increased
survival of very small infants, retinopathy of prematurity
has become one of the leading causes of childhood
blindness. Over the past two decades, major advances
have been made in understanding the pathogenesis of
ROP, to a large extent as a result of changes in clinical risk
factors (oxygen and non-oxygen related) and
characteristics observed in ROP cases. This article
provides a literature review on the evolution in clinical
characteristics, classification and treatment modalities and
indications of ROP. Special attention is hereby paid to the
neonatal factors influencing the development of ROP and
to the necessity for everyone caring for premature babies
to have a well-defined screening and treatment protocol
for ROP. Such screening protocol needs to be based on a
unit- specific ROP risk profile and, consequently, may vary
between different European regions.
2.2 An international classification of retinopathy
of prematurity
Because of modern life-support systems capable of
keeping tiny premature infants alive, retinopathy of
prematurity has recurred. No classification system
currently available adequately describes the observations
of the disease being made today. A new classification
system, the work of 23 ophthalmologists from 11
countries, is presented in an attempt to meet this need. It
emphasizes the location and the extent of the disease in
the retina as well as its stages
2.3 Automated detection of diabetic retinopathy
on digital fundus images
Diabetic retinopathy (DR) is a condition where the retina
is damaged due to fluid leaking from the blood vessels
into the retina. In extreme cases, the patient will become
blind. Therefore, early detection of diabetic retinopathy is
crucial to prevent blindness. The main stages of diabetic
retinopathy are non-proliferate diabetes retinopathy
(NPDR) and proliferate diabetes retinopathy (PDR). In
this study, we propose a system for automated
classification of normal, and abnormal retinal images by
automatically detecting the blood vessels, hard exudates
micro aneurysms, entropy and homogeneity. The
objective measurements such as blood vessels area,
exudates area, micro aneurysms area, entropy and
homogeneity are computed from the processed retinal
images. These objective measurements are finally fed to
the artificial neural network (ANN) classifier for the
automatic classification. Different approaches for image
restoration are tested and compared on Fundus images.
The effect of restoration on the automatic detection
process is investigated in this paper
2.4 U-net: Convolutional networks for
biomedical image segmentation
There is large consent that successful training of deep
networks requires many thousand annotated training
samples. In this paper, we present a network and training
strategy that relies on the strong use of data
augmentation to use the available annotated samples
more efficiently. The architecture consists of a contracting
path to capture context and a symmetric expanding path
that enables precise localization. Using the same network
trained on transmitted light microscopy images (phase
contrast and DIC) we won the ISBI cell tracking challenge
2015 in these categories by a large margin. Moreover, the
network is fast. Segmentation of a 512x512 image takes
less than a second on a recent GPU.
3. PROPOSED WORK FOR RETINOPATHY
In this section, the proposed system for retinopathy is
described. Classifying retinal disease based on machine
learning technique “K-nearest neighbor algorithm”.

Segmenting colors from different types for retinopathy
and classifying it using feature extraction, perimeter
finding and train the machine using KNN algorithm.
Testing the accuracy of a disease using confusion matrix
and classify the disease with above 97% accuracy.
A novel architecture of convolutional neural networks
(CNN) is proposed to recognize the existence and severity
of ROP. The architecture is composed of a feature extract
sub-network, followed by a feature aggregate operator to
bind features from variable images in an examination. The
prediction is accomplished using a second sub-network
with the aggregated features as inputs. Max and mean
aggregate operators are explored based on the
architecture. Several ImageNet pertained networks are
tested in the study, including VGG-16, Inception- V2, and
ResNet50. The proposed architecture is verified with a
large dataset of 2668 examinations of the fungus in infants.
The experimental results demonstrate that the Inception-
V2 with the max aggregate operator in module 2 is a
proper network architecture for the recognition of the
existence and severity of ROP. Compared with the mean
aggregate operator, the max has better classification
accuracy and convergence speed. Meanwhile, a patient’s
multiple examinations in train, validation, and test
datasets has little impact on model’s performance, mainly
because the characteristics of the eyes of the premature
infants are varied overtime.
3.1 System architecture
The input images with certain features, with all of those in
the training data set, the model comes to a decision by
judging the image with the severity of KNN and produce
training image features
3.2 Disadvantages of existing system
 CNN gives many frameworks.
 Only predict the stages of retinopathy of
prematurity
 Processor should be high configuration
 CNN has problem of over fitting and its mostly
computationally expensive because it has to take
a large database set for training
4. IMPLEMENTATION
4.1Image Segmentation
Segmentation is the process of assigning a label to every
pixel. In other words, the segmentation is partitioning a
digital image into multiple segments "pixels." The goal of
segmentation is to simplify the representation of an image
into something that is more meaningful and easier to
analyze. Whereas the result of image segmentation is a set
of segments that cover the entire image, or a set of
contours extracted from the image. The simplest method
of image segmentation is called the threshold method.
This method is based on threshold value to turn a gray
scale image into a binary image. During this process,
every pixel in an image is called as object pixel if the value
is greater than the threshold value and it is named as
background pixel if the value is lower than the threshold
value. An object pixel is being given a “1” value while the
background pixel is given the “0” value. After which a
binary image is being created with all the object and
background pixels.
4.2 Feature Extraction
Feature extraction starts from an initial set of measured
data and builds derived values (features) intended to be
informative and non-redundant, facilitating the
subsequent learning and generalization steps, and in
some cases leading to better human interpretations.
Feature extraction is related to dimensionality reduction.
When the input data to an algorithm is too large to be
processed and it is suspected to be redundant (e.g. the
same measurement in both feet and meters, or the
repetitiveness of images presented as pixels), then it can
be transformed into a reduced set of features (also named
a feature vector). Determining a subset of the initial
features is called feature selection.[1] The selected
features are expected to contain the relevant information
from the input data, so that the desired task can be
performed by using this reduced representation instead
of the complete initial data
4.3 Dataset Preparation:
A data set (or dataset) is a collection of data. In the case of
tabular data, a data set corresponds to one or more
database tables, where every column of a table represents
a particular variable, and each row corresponds to a given
record of the data set in question.
In the open data discipline, data set is the unit to measure
the information released in a public open data repository.
The European Open Data portal aggregates more than half

a million data sets.[2] In this field other definitions have
been proposed,[3] but currently there is not an official
one. Some other issues (real-time data sources,[4] non-
relational data sets, etc.) increases the difficulty to reach a
consensus about it.
4.4 Import the Dataset
A lot of datasets come in CSV formats. We will need to
locate the directory of the CSV file at first (it’s more efficient
to keep the dataset in the same directory as your program)
and read it using a method called read_csv which can be
found in the library called pandas.
After inspecting our dataset carefully, we are going to
create a matrix of features in our dataset (X) and create a
dependent vector (Y) with their respective observations. To
read the columns, we will use iloc of pandas (used to fix the
indexes for selection) which takes two parameters — [row
selection, column selection].
4.5 Taking Care of Missing Data in Dataset
Sometimes you may find some data are missing in the
dataset. We need to be equipped to handle the problem
when we come across them. Obviously you could remove
the entire line of data but what if you are unknowingly
removing crucial information? Of course we would not
want to do that. One of the most common idea to handle the
problem is to take a mean of all the values of the same
column and have it to replace the missing data.
The library that we are going to use for the task is called
Scikit Learn preprocessing. It contains a class called
Imputer which will help us take care of the missing data.
From sklearn .preprocessing import Imputer
A lot of the times the next step is to create an object of the
same class to call the functions that are in that class. We will
call our object imputer. The Imputer class can take a few
parameters —
i. missing values — We can either give it an integer or
“NaN” for it to find the missing values.
ii. strategy — we will find the average so we will set it to
mean. We can also set it to median or most_frequent (for
mode)as necessary.
iii. axis — we can either assign it 0 or 1, 0 to impute along
columns and 1 to impute along rows.
imputer = Imputer(missing_values = "NaN", strategy =
"mean", axis = 0)
4.6 Encoding Categorical Data
Sometimes our data is in qualitative form, that is we have
texts as our data. We can find categories in text form. Now
it gets complicated for machines to understand texts and
process them, rather than numbers, since the models are
based on mathematical equations and calculations.
Therefore, we have to encode the categorical data.
This is an example of categorical data. In the first column,
the data is in text form. We can see that there are five
categories — Very, Somewhat, Not very, Not at all, Not sure
— and hence the name categorical data.
So the way we do it, we will import the scikit library that
we previously used. There’s a class in the library called
LabelEncoder which we will use for the task from sklearn.
Preprocessing import LabelEncoder
As I have mentioned before, the next step is usually to
create an object of that class. We will call our object
labelencoder_X.
labelencoder_X = LabelEncoder()
4.7 Splitting the Dataset into Training Set and
Test Set
Now we need to split our dataset into two sets — a
Training set and a Test set. We will train our machine
learning models on our training set, i.e our machine
learning models will try to understand any correlations in
our training set and then we will test the models on our
test set to check how accurately it can predict. A general
rule of the thumb is to allocate 80% of the dataset to
training set and the remaining 20% to test set.
4.8 Feature Scaling
It is a method used to standardize the range of
independent variables or features of data.
But why is it necessary? A lot of machine learning models
are based on Euclidean distance. For example, the values in
one column (x) is much higher than the value in another
column (y), (x2-x1) squared will give a far greater value
than (y2-y1) squared. So clearly, one square difference
dominates over the other square difference. In the machine
learning equations, the square difference with the lower
value in comparison to the far greater value will almost be
treated as if it does not exist. We do not want that to
happen. That is why it is necessary to transform all our
variables into the same scale. There are several ways of
scaling the data. One way is called Standardization which
may be used.

4.9 K-Nearest Neighbor Algorithm in Python
k-Nearest Neighbors, or KNN for short, is one of the
simplest machine learning algorithms and is used in a wide
array of institutions. KNN is a non-parametric, lazy learning
algorithm. When we say a technique is non-parametric, it
means that it does not make any assumptions about the
underlying data. In other words, it makes its selection based
off of the proximity to other data points regardless of what
feature the numerical values represent. Being a lazy
learning algorithm implies that there is little to no training
phase. Therefore, we can immediately classify new data
points as they present themselves.
4.10 K- NEAREST NEIGHBOUR
knn=KNeighborsClassifier(n_neighbors=5,
metric='euclidean')
knn.fit(X_train, y_train)
y_pred = knn.predict(X_test)
KNN Algorithm
Load the data
Initialize K to your chosen number of neighbors
3. for each example in the data
3.1 Calculate the distance between the query example and
the current example from the data.
3.2 Add the distance and the index of the example to an
ordered collection
4. Sort the ordered collection of distances and indices from
smallest to largest (in ascending order) by the distances
5. Pick the first K entries from the sorted collection
6. Get the labels of the selected K entries
7. If regression, return the mean of the K labels
8. If classification, return the mode of the K labels
Advantages
The algorithm is simple and easy to implement.
There’s no need to build a model, tune several parameters,
or make additional assumptions.
The algorithm is versatile. It can be used for classification,
regression, and search (as we will see in the next section).
4.11 Flowchart :
5. RESULTS AND DISCUSSION
In this paper, we have used K-nearest neighbor algorithm
with accuracy of 97% to classify the different types of
retinopathy using the concept of machine learning.
Feature extracted from the image with the help of Huge
Saturated Value .This is followed by differentiating the
colors present in the image. Finally the Retinopathy is
classified .
Fig-2: Feature extraction of diabetic image
6. FUTURE PERSPECTIVE:
Future treatment will probably involve early vitreous
sampling followed by an injection of multiple combined
medications tailored to the individual's vitreous signature.
Use of risk profiles will facilitate improvements in risk
factor control, reduce incidence and progression of DR, and
reduce costs. In addition, manipulation of pharmacokinetic
drug properties may allow for a longer intravitreal half-life
and less need for frequent injections. As the duration and

efficacy of intravitreal drugs improve, the use of treatment
modalities that destroy retinal tissue, such as focal laser
photocoagulation or pan-retinal photocoagulation, may no
longer be necessary.
7. CONCLUSION
We have tried to construct an ensemble to predict if a
patient has retinopathy using features from retinal photos.
After training and testing the model the accuracy we get is
quite similar. Datasets is providing higher accuracy rate for
predicting retinopathy. Despite the shortcomings in
reaching good performance results, this work provided a
means to make use and test multiple machine learning
algorithms and try to arrive to ensemble models that would
outperform individual learners. It also allowed exploring a
little feature selection, feature generation, parameter
selection and ensemble selection problems and experiences
the constraints in computation time when looking for
possible candidate models in high combinatorial spaces,
even for a small dataset as the one used. The structure of
our research has been built in such a way that with proper
dataset and minor alternation it can work to classify the
disease in any number of categories.
REFERENCES
[1] B. C. Chu, and I. Y. Wong, “Incidence and risk factors for
retinopathy of prematurity in multiple gestations: A chinese
population study,”
Medicine, vol. 94, no. 18, pp. 185–191, 2015
[2] A. Gschlieer, E. Stifter, T. Neumayer, E. Moser, A. Papp, N.
Pircher, G. Dorner, S. Egger, N. Vukojevic, and I. Oberacher-
Velten, “Inter-expert and intra-expert agreement on the
diagnosis and treatment of retinopathy of prematurity,”
American Journal of Ophthalmology, vol. 160, no. 3, pp.
553–560, 2016.
[3] D. E. Worrall, C. M. Wilson, and G. J. Brostow,
“Automated retinopathy of prematurity case detection with
convolutional neural networks,” in International Workshop
on Large-Scale Annotation of Biomedical Data and Expert
Label Synthesis, 2016, pp. 68–76.
[4] N. Tajbakhsh, J. Y. Shin, S. R. Gurudu, R. T. Hurst, C. B.
Kendall, M. B. Gotway, and J. Liang, “Convolutional neural
networks for medical image analysis: Full training or fifine
tuning,” IEEE Transactions on Medical Imaging, vol. 35, no.
5, pp. 1299–1312, 2016.
[5] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D.
Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich,
“Going deeper with convolutions,” in Proceedings of the
IEEE Conference on Computer Vision and Pattern
Recognition, 2017.

IRJET - Classification of Retinopathy using Machine Learning

More Related Content

What's hot (20)

Similar to IRJET - Classification of Retinopathy using Machine Learning (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET - Classification of Retinopathy using Machine Learning