International Journal of Computer Trends and Technology (IJCTT) – Special Issue April - 2017

Disease Prediction Based On Retinal Images

Using Deep Neural Networks
1
T. Johnlayoni , 2R. Praseela ,3P. Ragavi , 4B. Dhivya
1,2,4
Student, Department of computer Science and Engineering,
4
Assistant Professor, Department of Computer Science and Engineering,
Arasu Engineering College, Kumbakonam, TamilNadu, India.

ABSTRACT--- Retina is responsible for capturing has been combined with scanning laser
the visual and it triggers the nerve impulses in the ophthalmoscope and optical coherence tomography
brain. The eye is sometimes said to provide a (OCT) to obtain images of retinal
window into the health of a person. There are a microvasculature and blood flow and three
number of diseases, particularly vascular disease dimensional images of living cone photoreceptors
that leave telltale markers in the retina. It enable respectively. opening which consisted of erode
optometrist curve to capture retinal blood supply. followed by dilate is applied first. Erode
Retinal images for identifying and quantifying the function protects the small blood vessels by
effects of diseases such as cardio vascular diseases. reducing their sizes while dilate function blows
A retinal image provides a information about what up the larger remaining. The automatic analysis of
is happening inside the human body. In particular, retinal fundus, a number of algorithms have been
the state of the retinal vessels has been shown to proposed for extracting the vascular structure and
reflect the cardiovascular condition of the body. So for identifying non-vascular
in this project, we can implement based on neural lesion (exudates, haemorrhages, ischemic
network method to provide regional information regions). The first changes in the retina that point
about arteries and veins. And finally predict cardio out the onset of a retinopathy, e.g. from a
vascular diseases and other diseases using CRAE systemic disease, appear in the vessels. Changes
and CRVE measurements. in vessel structure can affected.

Keywords- Vessel segment , CRAE, CRVE, Deep III. METHODOLOGIES:

Neural networks, segmentation, SVM Examination of blood vessels in the eye allows
detection of eye diseases such as glaucoma and
I. INTRODUCTION: diabetic retinopathy. Traditionally, the vascular
Eye is an important of human body. Retina of network is mapped by hand in a time-consuming
eye can provide variable information about human process that requires both training and skill.
health. Retinal image processing is greatly required Automating the process allows consistency, and
in diagnosing and treatment of many diseases most importantly, frees up the time that a skilled
affecting the retina. Retinal imaging is a recent technician or doctor would normally use for manual
technological advancement in eye care. It enables screening. So we can implement automatic process
optometrist to capture a digital image of the retina, to examine the blood vessels to identify the cardio
blood vessels and optic nerve located at the back of vascular diseases and other diseases in retinal
eyes. images. The proposed method utilizes the concept of
active contours to remove noise, enhance the image,
II. DESCRIPTION: track the edges of the vessels, calculate the width of
Retinal images obtained using Adaptive Optics vessels and identify diseases. Implement deep neural
have the potential to facilitate early detection of network algorithm to segment blood vessels and
retinal pathologies. Many researchers were working calculate width of the blood vessels. Finally
on retinal images to perform various image proposed approach provides vessel segmentation
processing tasks for the beneficial of health sector. with good performance. This will be a powerful tool
The result of image analysis relies on a preliminary for analyzing vasculature for better management of
phase of identifying good quality images, which a wide spectrum of vascular-related diseases. And
have high contrast have proved the automatic then implement support vector machine algorithm to
assessment of quality of retinal images taken by classify the retinal diseases using width values of
retinal images. Finally categorize the vessels as
fundal camera with a reference image. Recently, AO artery and vein vessels. Retinal vascular caliber

ISSN: 2231 - 2803 http://www.ijcttjournal.org Page 65

 International Journal of Computer Trends and Technology (IJCTT) – Special Issue April - 2017

(CRAE and CRVE) was analyzed as continuous

variables. We used analysis of covariance to
estimate mean retinal vascular caliber associated
with the presence versus absence of categorical
variables or increasing quartiles of continuous
variables to predict the cardio vascular and other
diseases.

IV. TECHNIQUES AND ALGORITHMS:

The segmented vessels are classified into
arteries and veins. Correct classification of vessels is
vital, because heart diseases affect arteries and veins
differently. The alterations in veins and arteries
cannot be analyzed without distinguishing them.
Segmented vessels are classified by the supervised
method Support Vector Machine. After extraction of
blood vessels, feature vector is formed based on
properties of artery and veins. The features get
extracted on the basis of centerline extracted image
and a label is assigned to each centerline, indicating CONCLUSION:
the artery and vein pixel. Based on these labeling We conclude that, our proposed system
phase, the final goal is now to assign one of the implemented successfully with accurate
labels with the artery class (A), and the other with identification of true vessels to obtain correct retinal
vein class (V). In order to allow the final ophthalmology measurements. And we implement
classification between A/V classes along with vessel the post processing step to vessel segmentation. This
intensity information the structural information and step is used to track all true vessels and find the
are also used. This can be done using SVM optimal forest. We can overcome wrong diagnosis
classification. The trained classifier is used for of crossovers by using simultaneous identification
assigning the A/V classes to each one of the sub of the significance of using structural information
graph labels. First, each centerline pixel is classified for A/V classification. Later vessel segmentation, it
into A or V classes, then for each label (Ci j , j = 1, is possible to achieve more progressive analysis,
2) in sub graph i, he probability of its being an such as measurements of diameters and lengths of
artery is calculated based on the number of the vessels, classification of veins and arteries,
associated centerline pixels classified by LDA to be calculation of arteries venous ratio, and more
an artery or a vein. The probability of label Ci j to prominently study the analytical and predictive
be an artery is Pa(Ci j )= naCi j /(naCi j + nvCi) values of these features on eye disease and a number
Where naCi j is the number of centerline pixels of a of systematic diseases. Furthermore, we compared
label classified as an artery and nvCi j is the the preface of our approach with other recently
number of centerline pixels classified as a vein. proposed methods, and we conclude that we are
For each pair of labels in each sub graph, the label achieving better results.
with higher artery probability will be assigned as an
artery class, and the other as a vein class. Finally, to
prevent a wrong classification as a result of a wrong
graph analysis, we calculate the probability of being
an artery or a vein for each link individually. REFERENCES:
[1]. K. Fukushima, “Neocognitron: A self- organizing
neural network model for a mechanism of pattern
recognition unaffected by shift in position,”
Biological cybernetics, vol. 36, no. 4, pp. 193–202,
1980.
[2]. Y. LeCun, B. Boser, J. S. Denker, D. Henderson,
R. E. Howard, W. Hubbard, and
[3]. L. D. Jackel, “Backpropagation applied to
handwritten zip code recognition,” Neural
computation, vol. 1, no. 4, pp. 541–551,1989. [3] B. B.
LeCun, J. S. Denker, D. Henderson, R. E. Howard, W.
Hubbard, and L. D. Jackel, “Handwritten digit
recognition with a back- propagation network,” in
V. ARCHITECTURE DIAGRAM: Advances in neural information processing systems.
Citeseer, 1990.

ISSN: 2231 - 2803 http://www.ijcttjournal.org Page 66

 International Journal of Computer Trends and Technology (IJCTT) – Special Issue April - 2017

A. Krizhevsky, I. Sutskever, and G. E.

Hinton, “Imagenet classification with
deep convolutional neural networks,” in
Advances in neural information
processing systems, 2012, pp. 1097–1105.
[4]. X. Glorot, A. Bordes, and Y. Bengio, “Deep
sparse rectifier neural networks,” in International
Conference on Artificial Intelligence and Statistics,
2011, pp. 315– 323.
[5]. N. Srivastava, G. Hinton, A. Krizhevsky,
[6]. Sutskever, and R. Salakhutdinov, “Dropout: A
simple way to prevent neural networks from
overfitting,” The Journal of Machine Learning
Research, vol. 15, no. 1, pp. 1929–1958, 2014.
[7]. Y. Bengio, P. Lamblin, D. Popovici, H. Larochelle
et al., “Greedy layer-wise training of deep
networks,” Advances in neural information
processing systems, vol. 19, p. 153, 2007.
[8]. M. Fraz, P. Remagnino, A. Hoppe, B.
Uyyanonvara, A. Rudnicka, C. Owen, and S. Barman,
“Blood vessel segmentation methodologies in retinal
images - a survey,” Comput. Methods
Prog.Biomed.vol. 108, no. 1, pp. 407–433, Oct.
2012. [Online].
[9]. Available:
http://dx.doi.org/10.1016/j.cmpb.2012.03.00 9
[10]. J. Staal, M. D. Abràmoff, M. Niemeijer, M. A.
Viergever, and B. van Ginneken, “Ridge-based
vessel segmentation in color images of the retina,”
Medical Imaging, IEEE Transactions on, vol. 23,
no. 4, pp. 501–509, 2004.
[11]. A.Hoover, V. Kouznetsova, and M. Goldbaum,
“Locating blood vessels in retinal images by
piecewise threshold probing of a matched filter
response,” Medical Imaging, IEEE Transactions on,
vol. 19, no. 3, pp. 203–210, 20

ISSN: 2231 - 2803 http://www.ijcttjournal.org Page 67