Professional Documents
Culture Documents
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
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.
A number of researches were made in the area of diabetic retinopathy detection. Some of
them are explained below.
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.
CHAPTER 3
6
PROPOSED ALGORITHM
Retinal hemorrhage
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
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.
12
Pixel count
Mean
Standard deviation
13
Examples
A = [1 2 3; 3 3 6; 4 6 8; 4 7 7];
mean(A)
ans =
3.0000
4.5000
6.0000
14
15
16
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
CHAPTER 4
FLOW CHART
4.1 FLOW CHART FOR TRAINING
START
FEEDING THE
IMAGES
IMAGE CROPPING
IMAGE
CONVERSION
FEATURE
EXTRACTION
NETWORK
TRAINING
STOP
18
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
YES
NO
DISPLAY OUTPUT AS
EXUDATES
19
STOP
DISPLAY OUTPUT AS
NORMAL
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
Mean
Standard deviation
Output
<150
<25
Normal
>150
>25
Exudates
21
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 goal is achieved in 5 epochs, then the performance characteristics will be as
shown below.
CHAPTER 6
CONCLUSION
23
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
6.2 DISADVANTAGES
Costly implementation.
Limitation of choosing the size of the input image.
This study can be extended for the analysis of other diseases like hypertension, stroke,
migraine, and hearing loss.
APPENDIX
PROGRAM CODE
24
round(A);
disp(A)
if A>1
disp('Exudate');
else
disp('Normal');
end
end
REFERENCES
29
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
30