Project Report 2016

Diabetic Retinopathy Detection

CHAPTER 1
INTRODUCTION
1.1 DIABETIC RETINOPATHY
Diabetic retinopathy is one of the common complications of diabetes. Unfortunately, in
many cases the patient is not aware of any symptoms until it is too late for effective treatment.
The screening of diabetic patients for the development of diabetic retinopathy potentially reduces
the risk of blindness in these patients by 50%. It has been estimated that 30,000 individuals per
million total populations need to be examined to implement such a program. This is beyond the
scope of currently existing ophthalmology departments and would produce a heavy clinical
burden if left to diabetic physicians. Photographing the fundus of such patients with later
assessment of the photographs has been tried with some success, but still requires the relatively
expensive time of the specialist ophthalmologist to provide a classification of the retinopathy.
This may delay referral of the patient for further examination. A wholly automated approach
involving fundus image analysis by computer could improve the efficiency of the assessment of
the image by providing an immediate classification of the fundus of the patient at the time of
acquisition of the image.
Retinal hemorrhage is a disorder of the eye in which bleeding occurs into the retina. The
retina is a thin disc-shaped layer of light-sensitive tissue on the back wall of the eye. Its job is to
translate what we see into neural impulses and send them to the brain via the optic nerve. A
retinal hemorrhage can be caused by hypertension, retinal vein occlusion (a blockage of a retinal
vein), or diabetes mellitus (which causes small fragile blood vessels to form, which are easily
damaged).Diabetic retinopathy is a common eye problem associated with diabetes. Diabetes, by
stressing the circulatory system, can cause damage, including hemorrhaging, to the small blood
vessels of the retina.On-proliferative retinopathy occurs when the damaged or leaking blood
vessels do not spreading. Proliferative retinopathy occurs when new blood vessels begin to form
in damaged areas of the retina and may lead to spots, floaters, decreased vision, or sudden loss of
vision. Sudden vision loss may occur if one of the newly formed blood vessels ruptures. Due to
increased pressure in the area, the retina may detach from the back of the eye, a serious condition
and a cause of blindness
1

Project Report 2016

Diabetic Retinopathy Detection

Exudates is a fluid with a high content of protein and cellular debris which has escaped
from blood vessels and has been deposited in tissues or on tissue surfaces, usually as a result of
inflammation. Hard exudates (Intra-retinal lipid exudates) are yellow deposits of lipid and
protein within the sensory retina. Accumulations of lipids leak from surrounding capillaries and
micro aneurysms. The occurrence of diabetic retinopathy will result in the disturbance of visual
capability and can eventually lead to blindness. Mainly there are two disease conditions may
cause this, they are exudates and hemorrhage.
Diabetic retinopathy isretinopathy (damage to the retina) caused by complications of
diabetes mellitus, which can eventually lead to blindness. It is an ocular manifestation of
systemic disease which affects up to 80% of all patients who have had diabetes for 10 years or
more. Despite these intimidating statistics, research indicates that at least 90% of these new cases
could be reduced if there was proper and vigilant treatment and monitoring of the eyes.Diabetic
retinopathy often has no early warning signs. As new blood vessels form at the back of the eye as
a part of proliferative diabeticretinopathy (PDR), they can bleed (ocular hemorrhage) and blur
vision. The first time this happens, 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 vision.
Diabetic retinopathy is the most common diabetic eye disease and a leading cause of
blindness in American adults. It is caused by changes in the blood vessels of the retina. In some
people with diabetic retinopathy, blood vessels may swell and leak fluid. In other people,
abnormal new blood vessels grow on the surface of the retina. The retina is the light-sensitive
tissue at the back of the eye. A healthy retina is necessary for good vision. If you have diabetic
retinopathy, at first you may not notice changes to your vision. But over time, diabetic
retinopathy can get worse and cause vision loss. Diabetic retinopathy usually affects both eyes.
Our project work is aimed to develop an automated system to analyze the retinal images
for important features of diabetic retinopathy using image processing techniques and an image
classifier based on artificial neural network which classifies the images according to the disease
conditions.

Project Report 2016

Diabetic Retinopathy Detection

A number of researches were made in the area of diabetic retinopathy detection. Some of
them are explained below.

1.2 AUTOMATIC RECOGNITION OF RETINOPATHY DISEASES BY

USING WAVELET BASED NEURAL NETWORK
The method used for the study is the retinal images were resized and pre-processed. In
this work the wavelet transform was used as a method of feature extraction before classifying the
retinal images. Then the resized images are transformed to gray scale. Then the wavelet
coefficients are found by using matlab wavelet toolbox. These features are applied as the input to
the neural networks.

Figure 1.1: Block diagram of Automatic Recognition of Retinopathy

The objective of this study is to investigate a good method for automated analysis of
retinal images for the purpose of detecting and recognition retinopathy diseases. The results are
very good so far and show that proposed method is very useful and help medical doctors to early
diagnosis the retinopathy diseases. In this study we used 3 images each Patients retinopathy
diseases of in the data set for training the neural network and the performance should be
improved by using more. This proposed method can recognize with retinopathy accurately, in
comparison to other methods and is potentially a powerful tool for the recognition of retinopathy
diseases.

Project Report 2016

Diabetic Retinopathy Detection

1.3 AUTOMATICDETECTION OF HARD EXUDATES IN

DIABETICRETINOPATHYUSINGMORPHOLOGICAL SEGMENTATION
ANDFUZZY LOGIC
In this paper it is presented a novel approach to automatically detect the presence of
Diabetic Retinopathy in color digital retinal images. The proposed approach utilizes the
morphological operations for the segmentation and fuzzy logic for the identification of features
of diabetic retinopathy in digital fundus images. This approach utilizes the features like hard
exudates, soft exudates and the red lesions such as Micro aneurysm, hemorrhages of diabetic
retinopathy to detect the presence of it in retinal images.
The digital retinal images are segmented using the morphological operations to identify
the regions showing signs of diabetic retinopathy. The XYZ, YIQ, LUV,HSVand Lab color space
of the identified regions are determined and a fuzzy set is formed from the values. Then fuzzy
rules are derived from the fuzzy set based on fuzzy logic. These fuzzy rules are used to detect the
presence of diabetic retinopathy in digital fundus images. We have used the publicly available
dataset DIARETDB0 (Standard Diabetic Retinopathy dataset) for the evaluation of the proposed
algorithm.

Project Report 2016

Diabetic Retinopathy Detection

CHAPTER 2
PROBLEM IDENTIFICATION AND THE PROPOSED
SOLUTION.
Several different computational strategies have been used in efforts to solve the
problem of automated diabetic retinopathy. These attempts have been limited both by the amount
of data available to researchers in this area and in the variety of methods used to solve the
problem. A common theme in this literature of this area has been the need to split the problem up
into first identifying the normal features or parts of the retina such as blood vessels, fovea and
optic disc, then attempting to identify and possibly localize exudates and hemorrhage versus
blood vessels and exudates versus optic disc. In certain image doing so is challenging even for
trained professionals.
While these methods have had varying success identifying and localizing components
of the retina, they typically operate on images with a good amount of contrast, few occlusions of
retinal objects and few if any manifestations of retinal disease. The last 15 years has been
however a steady increase in the literature that attempts to tackle the problem of detecting not
only retinal components in diseased eyes, but also retinopathy components. Generally speaking,
one may split these methods into morphological methods include filtering and segmentation.
In one analysis, first the input features based on characteristics of exudates bright area, closely
cluster white or yellowish color and strong edge were selected. Then blood vessels and optic disc
pixels were removed from the images in order prevent misclassification by pre-processing.
However the algorithm still has some false detection because some pixels with similar color to
the exudates belong to optic disc and edge of blood vessels.
A number of attempts have been made to use machine learning to automatically locate
manifestation of retinopathy. Examples includes unsupervised methods such as PCA (Li and
Chutatape 2003), k-means clustering and Gaussian mixture model. Overall most of these
attempts have come from the medical imaging community. There remain however many
techniques from recent computer vision literatures that have not been tried. Conversely, many of
the computer vision based attempts seem inefficient in practical terms because they do not
generalize well when presented with the variety of real data observed by practitioners.

Project Report 2016

Diabetic Retinopathy Detection

2.1 PROPOSED SOLUTION

The existing algorithm for the detection of diabetic retinopathy has some false
detection because some pixels with similar color to the exudates belong to optic disc and edge of
blood vessels. Also the existing methods have had varying success identifying and localizing
components of the retina, since they typically cannot operate on images with a less amount of
contrast.
In order to eliminate the problems of the existing system for the early detection of
diabetic retinopathy such as false detection, we are proposing a new approach. In our method for
detecting the diabetic retinopathy the scope of the digital image processing and artificial neural
networking is utilized. Here we can process a retinal image having less amount of contrast since
the possibilities of digital image processing is used. That means a less contrast image can be preprocessed by digital image processing for making the input image a better one. Also the chance
of false detection can be minimized with the help of implementing the artificial neural networks.
Here we are using supervised two layer back propagation network in which the possibility of
occurrence of error can be minimized. In this work we developed an automated tool for the early
detection of diabetic retinopathy. For this the fundus photographs of retina were taken with a
fundus camera during mass screening. These photographs were then scanned by a flat bed
scanner and saved as image files. The image files were then analyzed using the algorithms.

CHAPTER 3
6

Project Report 2016

Diabetic Retinopathy Detection

PROPOSED ALGORITHM

Figure 3.1: Block diagram of the proposed algorithm

3.1 BLOCK DIAGRAM DISCRIPTION

3.1.1 INPUT IMAGES
For the image dataset digital fundus images are obtained using a special camera called
fundus camera. A fundus camera or retinal camera is a specialized low power microscope with
an attached camera designed to photograph the interior surface of the eye, including the retina,
optic disc, macula, and posterior pole (i.e. the fundus). Fundus cameras are used by optometrists,
ophthalmologists, and trained medical professionals for monitoring progression of a disease,
diagnosis of a disease (combined with retinal angiography), or in screening programs, where the
photos can be analyzed later A fundus camera provides an upright, magnified view of the fundus.
A typical camera views 30 to 50 degrees of retinal area, with a magnification of 2.5 xs

3.1.1.1 Fundus Camera Specifications.

7

Project Report 2016

Diabetic Retinopathy Detection

Focus Aid: Eye position dots

Monitor: Computer screen, Anterior LCD(optional)
Camera Resolution: Max 12 MEGA (72 pixels per degree minimum)
Field of view: 45 degree both horizontally and vertically
Working distance: 40mm
Vertical tilting: 15
Horizontal tilting: 30
Chin rest distance: 70mm
Electric-table distance: 200mm
Light adjustment: Inching and coarse stepless adjustment
Fixation lamp: Red LED
Working lamp: 12V 60W
Flash lamp:
Power supply: AC 220V 50Hz
Weight: 50kg
Software specifications
* Image file formats: Default. BMP and. JPG, the. JPG image is compressed max to 50:
1, both the formats do not result in the loss of any clinical significant.

Project Report 2016

Diabetic Retinopathy Detection

Figure 3.2: Image of fundus camera

The fundus photographs were taken during mass screening. These photographs were
then scanned by a flat bed scanner and saved as image files. The image files were then analyzed.
The input retinal fundus image can be of three types, they are

Normal retinal image

Retinal hemorrhage image

Retinal exudates image

Project Report 2016

Diabetic Retinopathy Detection

Normal retinal image

Retinal hemorrhage

Retinal exudates image

Figure 3.3: Retinal images

All the input retinal images are RGB or color images. The images we have selected for
the disease identification generally have a size of 250*200, and each image is represented in the
form of a matrix having dimension of 250*200 ( 250 columns and 200 rows). The matrix
representation of input images actually consists of three sub matrices which correspond to the
value of Red, green, blue of each pixel in the original image. Thus each pixel of the input image
has a value which is the combination of corresponding three values in the matrix representing
R,G,B.

10

Project Report 2016

Diabetic Retinopathy Detection

Figure 3.4: Color RGB Photograph

3.1.2 IMAGE CROPPING

Cropping refers to the removal of the outer parts of an image to improve framing,
accentuate subject matter or change aspect ratio. A digital image is made up of pixels. Cropping
is the process of removing selected pixels from a digital image. The collected normal as well as
diseased retinal images are cropped in order to make them into equal sizes
For cropping operation to perform we have to specify the upper left corner and the
bottom right corner of the particular rectangular image. For example, suppose rect is [20 20 40
30], using the default spatial coordinate system. The upper-left corner of the specified rectangle
is the center of the pixel (20,20) and the lower-right corner is the center of the pixel (40,30).
11

Project Report 2016

Diabetic Retinopathy Detection

After cropping the matrix representing the cropped image will have smaller dimension. But the
values of the pixels in the cropped image will be same as that in the original image.

Original image

cropped image
Figure 3.5: Image cropping

In our project we have selected the cropping size as [50 45 150 120], which represents
the lens portion of retina as shown in the above figure.

3.1.3 FEATURE EXTRACTION

The images are analyzed to extract the features of normal retina, retinal hemorrhage and
retinal exudates. The feature set should be selected such that the between class discrimination is
maximized while the within class discrimination is minimized. Based on these criteria we have
selected three features for the disease identification. Tey are:

12

Pixel count

Project Report 2016

Mean

Standard deviation

Diabetic Retinopathy Detection

3.1.3.1 Pixel Count

In order to take the pixel count first we have to convert the cropped image to its
corresponding binary image. For converting to binary we have to fix a threshold value say 0.45.
Based on this threshold value the pixel can be either black or white. That is the pixels having
value greater than this threshold value will be assigned a binary 1 and those having value lesser
than the threshold value will be assigned as binary 0. Thus the matrix representing binary
image consist of only 1s and 0s.
As stated earlier hemorrhage is the breakage of blood vessels, so while examining the
retinal hemorrhage images it is come to know that most of the pixels are dark; and after
converting to binary it appears as black. That is the no of black pixels will be greater than the
number of black pixels. So we fix the black pixel count as the first feature for identifying the
disease condition. By examining about 30 retinal hemorrhage images it is observed that the pixel
count of these images is greater than 5000.

3.1.3.2 Mean And Standard Deviation

Mean is a measure of average gray level in an image and the standard deviation, which
is a measure of average contrast. Finding the mean and the standard deviation of an image, first
we have to convert that image into a grayscale. Grayscale images are distinct from one-bit black
and white images, which in the context of computer imaging are images with only the two colors
black and white. Grayscale images have many shades of gray in between zero and one.To
convert any color to a grayscale representation of its luminance, first one must obtain the values
of its red, green, and blue (RGB) primaries in linear intensity encoding.
If A is a vector, mean (A) returns the mean value of A. If A is a matrix, mean(A) treats
the columns of A as vectors, returning a row vector of mean values

13

Project Report 2016

Diabetic Retinopathy Detection

Examples
A = [1 2 3; 3 3 6; 4 6 8; 4 7 7];
mean(A)
ans =
3.0000

4.5000

6.0000

3.1.4 DISEASE IDENTIFICATION

The classification of input images is done using an artificial neural network. The
principle advantages of artificial neural network are that they are able to generalize, adapting to
signal distortion and noise without loss of robustness. They are trained by example and do not
require precise description of patterns to be classified or criteria for classification. Based on the
pixel count of the cropped image, mean and standard deviation the given input retinal images are
classified as normal, exudates and hemorrhage.
The network was trained to recognize features in the retinal image. In this work a 2layer supervised back propagation network is used. Two layers are 1.input layer 2. Output layer.
There are three neurons in the input layer. The criteria for selecting the number of neurons in the
input layer are the number of extracted features. They are pixel count, mean and standard
deviation. The hemorrhage images are already classified based on pixel count. Now we have to
classify the remaining two; they are exudates and normal. This is the reason for selecting two
layer networks. If we need more classification more layers can be included. These additional
layers are called hidden layers. For the training process we have to fix a target. Here the target is
a matrix of order 1x20 consisting of ten zeros and ten ones. The network is trained a number of
times until the error is minimized and goal is achieved.
To calculate the weight changes in the hidden layer the error in the output layer is backpropagated to these layers according to the connecting weights. This process is repeated for each
sample in the training set. One cycle through the training set is called an epoch. The number of
epochs needed to train the network depends on various parameters especially on the error
calculated in the output layer.

14

Project Report 2016

Diabetic Retinopathy Detection

3.1.4.1 Artificial Neural Network

ANN is a mathematical or computational model and consists of an interconnected
group of artificial neurons and processes information using a connectionist approach to
computation.
A single input neuron is shown in figure. The scalar input p is multiplied by the scalar
weight w to form wp , one of the terms that is sent to the summer. The other input l is
multiplied by a bias b and then passed to the summer. The summer output n, often referred to as
the net input, goes into a transfer function f, which produces the scalar neuron output a.
If we relate this simple model to the biological neuron, the weight w corresponds to
the strength of a synapse, the cell body is represented by the summation and the transfer
function, and the neuron output a represents the signal on the axon.

Figure 3.6: Single input neuron

The neuron output is calculated as

15

Project Report 2016

Diabetic Retinopathy Detection

a = f (wp + b)

if , for instance, w=3 , p=2 and b= -1.5, then

a = f (3(2) 1.5) = f (4.5)
the actual output depends on the particular transfer function that is chosen.
The bias is much like a weight, except that it has a constant input of 1. However, if you

3.1.4.2 Multilayer Neural Network

Multilayer networks solve the classification problem for non linear sets by employing
hidden layers, whose neurons are not directly connected to the output. The additional hidden
layers can be interpreted geometrically as additional hyper-planes, which enhance the separation
capacity of the network. Figure 2.2 shows typical multilayer network architectures. It do not
want to have a bias in a particular neuron, it can be omitted. Note that w and b are both
adjustable scalar parameters of the neuron. Typically the transfer function is chosen by the
designer and then the parameters w and b will be adjusted by some learning rule so that the
neuron input/output relationship meets some specific goal.
This new architecture introduces a new question: how to train the hidden units for
which the desired output is not known. The Backpropagation algorithm offers a solution to this
problem.
The training occurs in a supervised style. The basic idea is to present the input vector to
the network; calculate in the forward direction the output of each layer and the final output of the
network. For the output layer the desired values are known and therefore the weights can be
adjusted as for a single layer network; in the case of the BP algorithm according to the gradient
decent rule.

16

Project Report 2016

Diabetic Retinopathy Detection

Figure 3.7: Network layer

3.1.5 OUTPUT
We made the first disease identification that is hemorrhage for those images having
pixel count greater than 5000. The exudates and the normal images cannot be classified based on
pixel count because their pixels counts are almost similar. So we go for another two parameters
which are mean and standard deviation
By testing and training of image we compare the outputs of the output layer neurons
with the predetermined targets and if they match, corresponding output is displayed. The test
performance of the network is evaluated by computing the statistical parameters such as mean
and standard deviation which are given below.
While taking the mean value it is observed that for normal images the values are less
than 150, and for exudates it is greater than 150. In the case of standard deviation the values are
less than 25 for normal images and greater than 25 for exudates.

17

Project Report 2016

Diabetic Retinopathy Detection

CHAPTER 4
FLOW CHART
4.1 FLOW CHART FOR TRAINING

START

FEEDING THE
IMAGES

IMAGE CROPPING

IMAGE
CONVERSION

FEATURE
EXTRACTION
NETWORK
TRAINING

STOP

18

Project Report 2016

Diabetic Retinopathy Detection

Figure 4.1: Flow chart for training the neural network

4.2 FLOW CHART FOR TESTING

START
FEEDING IMAGES

IMAGE CROPPING
IMAGE CONVERSION
PIXEL COUNT
(C)COMPUTATION
SORT IMAGES ACCORDING TO
PIXEL COUNT
C >5000

YES

DISPLAY OUTPUT AS
HEMORRHAGE

NO
DETERMINE THE MEAN
&STANDARD DEVIATION
COMPUTATION
SIMULATION

COMPARE THE OUTPUT WITH

TARGET
O/P <1

YES

NO
DISPLAY OUTPUT AS
EXUDATES

19

STOP

DISPLAY OUTPUT AS
NORMAL

Project Report 2016

Diabetic Retinopathy Detection

Figure 4.2: Flow chart for testing

CHAPTER 5
RESULTS
The input retinal images are classified and disease conditions are identified as exudates
and hemorrhages or whether it is normal image based on three extracted features, pixel count,
mean and standard deviation. There are certain threshold values for these features in order to
make the disease identification more accurate.
For identifying the hemorrhage the threshold value is taken as 5000(pixel count). For
classifying normal images and exudates the mean and standard deviation are the extracted
features. The values of pixel count, mean and standard deviation of twenty images are given
below.

20

Project Report 2016

Diabetic Retinopathy Detection

Mean

Standard deviation

Output

<150

<25

Normal

>150

>25

Exudates

Table 5.1: values stored in p matrix

Based on the above values of mean and standard deviation the normal and exudates are
classified as follows.

21

Project Report 2016

Diabetic Retinopathy Detection

Table 5.2: Classification table

Consider an exudates retinal image, then its corresponding cropped, binary, grey scale
images will be appeared as shown below.

Figure 5.1: Output figure window

Suppose we fix the number of epochs as 1000 and goal as 0.01. The network will stop
training when one of the following conditions is satisfied.

22

When the maximum number of epochs is reached.

When the performance is minimized to goal.
Maximum amount of time is exceeded.

Project Report 2016

Diabetic Retinopathy Detection

When the goal is achieved in 5 epochs, then the performance characteristics will be as
shown below.

Figure 5.2: Performance characteristics

CHAPTER 6
CONCLUSION

23

Project Report 2016

Diabetic Retinopathy Detection

The eye diseases mainly contribute to blindness and often cannot be remedied because
the patients are diagnosed too late with the diseases. In this paper we described image processing
techniques, which can play a major role in the diagnosis of diabetic retinopathy. In this work the
neural network classifier is developed as an automated diagnostic tool to aid the physician in the
detection of these eye abnormalities. The accuracy achieved depends on various factors such as
the parameters used and the feature set. A system for classification of diabetic retinopathy using
digital image processing and artificial neural network has beendeveloped. This project work
describes a specific application,which can be extended to further applications in
medicine.Presently we are testing the system on a large patient offline database and in future it
can be implemented forroutine clinical use. This method of classification ofdiabetic retinopathy
condition using artificial neural networks almost coincides with expectedretinopathy condition.
These results will have significantusage in analyzing the diabetic retinopathy condition.This
system provides early warning of diabeticretinopathy abnormalities for diabetic patients.

6.1 ADVANTAGES

Early detection of DR will reduces the complication.

Reduction in significant amount of workload and time for ophthalmologists.

ANN is able to generalize since they are trained by example.

6.2 DISADVANTAGES

Costly implementation.
Limitation of choosing the size of the input image.

6.3 FUTURE SCOPE

This study can be extended for the analysis of other diseases like hypertension, stroke,
migraine, and hearing loss.

APPENDIX
PROGRAM CODE
24

Project Report 2016

PROGRAM FOR TRAINING

clc;
delete p;
for i=1:20;
I=num2str(i);
I1=strcat(I,'.jpg');
I2=imread(I1);
I3=imcrop(I2,[50 45 150 120]);
subplot(2,2,1);
imshow(I2)
subplot(2,2,2);
imshow(I3)
I4=im2bw(I3,.45);
subplot(2,2,3);
imshow(I4)
[j,k]=size(I4);
c(i)=0;
for j=1:120
for k=1:150
25

Diabetic Retinopathy Detection

Project Report 2016

if I4(j,k)==0
c(i)=c(i)+1;
end
end
end
if c>5000
disp('Hemorrhage');
else
I5=rgb2gray(I3);
subplot(2,2,4);
imshow(I5);
m(i)=mean2(I5);
s(i)=std2(I5);
end
end
p=[c;m;s];
disp(size(p));
p1=p';
t=[0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1];
26

Diabetic Retinopathy Detection

Project Report 2016

t1=t';
net=newff(p,t,[3,1],{'tansig','purelin'})
net.trainParam.epochs = 1000;
net.trainParam.goal = 0.01;
net=train(net,p,t)

PROGRAM FOR TESTING

i=input('enter the test image');
I=num2str(i);
I1=strcat(I,'.jpg');
I2=imread(I1);
I3=imcrop(I2,[50 45 150 120]);
subplot(2,2,1);
imshow(I2)
subplot(2,2,2);
imshow(I3)
I4=im2bw(I3,.45);
subplot(2,2,3);
imshow(I4)
[j,k]=size(I4);
27

Diabetic Retinopathy Detection

Project Report 2016

c=0;
for j=1:120
for k=1:150
if I4(j,k)==0
c=c+1;
end
end
end
if c>5000
disp('Hemorrhage');
else
I5=rgb2gray(I3);
subplot(2,2,4);
imshow(I5);
m=mean2(I5);
s=std2(I5);
p=[c;m;s];
y=sim(net,p);
A=10.*y;
28

Diabetic Retinopathy Detection

Project Report 2016

Diabetic Retinopathy Detection

round(A);
disp(A)
if A>1
disp('Exudate');
else
disp('Normal');
end
end

REFERENCES

29

Project Report 2016

Diabetic Retinopathy Detection

1. http://www.theberries.ns.ca/ARchives/2006Winter/diabetic_retinopathy.html
2. http://www.nei.nih.gov/health/diabetic/retinopathy.asp.
3. 2009IEEE International Advance Computing Conference (IACC 2009) Patiala, India, 67 March 2009 (Neural Computing Based Abnormality Detection in Retinal Optical
Images)
4. H Wang et al, An effective approach to detect lesions in color retinal images, IEEE
conference on computer vision and pattern recognition, pp: 181-187, 2000.
5. http://reseauophdiat.aphp.fr/Document/Doc/confliverpool.pdf#search='www.drsceening2
005.org
6. Xiaohui Zhang , A SVM Approach for Detection of Hemorrhages in Background

Diabetic Retinopathy, Proceedings of International Joint Conference on NeuralNetworks,

Montreal, Canada, July 31 - August 4, 2005.

30