IRJET- Glaucoma Detection using Convolutional Neural Network

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 01 | Jan 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 2078
Glaucoma Detection using Convolutional Neural Network
Aniket Patil1, Risha Shetty2, Sakshi Jain3, Sejal D’mello4
1,2,3
B.E. student, Dept. of Information and Technology, Atharva college of Engineering, Maharashtra, India
4
Assistant Professor, Dept. of Information and Technology, Atharva college of Engineering, Maharashtra, India
------------------------------------------------------------------------***-----------------------------------------------------------------------
Abstract –Glaucoma is a group of related eye disorders that cause damage to the optic nerve that carries information from the
eye to the brain which can get worse over time and lead to blindness. It is very important that glaucoma is detected as early as
possible for proper treatment. In this paper, we have proposed a Convolutional Neural Network (CNN) system for early detection
of Glaucoma. Initially, eye images are augmented to generate data for Deep learning. The eye images are then pre- processed to
remove noise using Gaussian Blur technique and make the image suitable for further processing. The system is trained using the
pre-processed images and when new input images are given to the system it classifies them as normal eye or glaucoma eye based
on the features extracted during training.
Key Words: Glaucoma, Convolutional Neural Network, Deep Learning, Gaussian Blur.
1. INTRODUCTION
Glaucoma is often linked to a build-up of pressure inside the eyes. Glaucoma tends to run in families and one usually doesn’t get it
until later in life. The increased pressure in eyes, called intraocular pressure, can damage the optic nerve, which sends images to the
brain. If the damage worsens, glaucoma can cause permanent vision loss or even total blindness within a few years. Most people with
glaucoma have no early symptoms or pain. One must visit the eye doctor regularly so they can diagnose and treat glaucoma before
one has long-term vision loss. If a person loses his vision, it can’t be brought back. But, lowering eye pressure can help keep the sight
that he has. Most people with glaucoma who follow their treatment plan and have regular eye exams are able to keep their vision. An
optic disc and cup are present in all individuals but an abnormal size of the cup with respect to the optic disc is a characteristic of a
glaucoma infected eye.
Traditional methods of detecting glaucoma include an eye doctor analysing the images and finding the abnormalities in it. This
method is very time consuming and not always accurate because the image contains noise and other factors which make it difficult
for proper analysis. Also, if a machine trained for analysis it becomes more accurate than human analysis.
Most of the literature works present mainly focus on optic cup and disk segmentation and some focus on Cup/Disk ratio. Through
our analysis we have found using Convolutional Neural Network model to be better than other literature works proposed.
Convolution neural networks are one of the most popular deep learning techniques for image analysis. In this technique, training data
sets of previously classified images are used to develop the system. This deep learning technique implies that computers can perform
feature learning and classification simultaneously. In deep learning algorithms, a model is formed using many layers which transform
the given input data to an output.
2. METHODOLOGY
Figure 1: Workflow

A. Database
The database contains total 100 eye images for training and testing. Out of 100 images, 50 images contain glaucoma and 50
images are healthy (without glaucoma). All the images are of varying sizes and in different formats.
B. Pre-processing
As mentioned above, all images in the dataset vary in size and format which may consist of noise in them. The image pre-
processing technique will help in reducing the noise and convert the data into a standard format for further classification of
images. Gaussian Blur is applied on the fixed size images(256x256 pixel) in order to reduce and remove noise.
C. Classification
The process of classification is to classify an image according to its visual content. The classifier used is k-nearest neighbour.
The number of neighbour (k), distance matrix and dataset is given as an input for the algorithm.
D. CNN Model
The pre-processed image is given as an input to the CNN model which consists of an input layer, convolution layers and a fully
connected layer. The input image of 256x256pixel acts as the input layer. In the first convolution layer, 16 filters of 3x3 size
kernels each are applied to the input image by gliding one by one through the position and a total of 16 feature maps are
generated. This method is called as feature extraction. These features are then applied to the ReLU activation function, which
performs a threshold operation for each input variable with values below zero. On the output of the ReLU layer, a max pooling
layer of 2x2 window size is applied which results in down-sampling ofthe feature maps to 128x128 pixel.
The output of the last convolution layer acts as an input for the next convolution layer. The next convolution layer contains of
32 3x3 scale kernel filters that are applied to each of function maps retrieved from the last layer. Specific operations like ReLU
and max pooling are carried out to generate 64x64 pixel down-sampled data.
The same operations are performed on the third layer which is the last layer where 64 filters of 3x3 size kernels are used which
produce 32x32 pixel data.
The third convolution layer has an output of 64 32x32 pixel feature maps. These features are then leveled to a single 32x32x64
= 65536 long vector, that acts as an input to a fully-connected layer. These features are then used to evaluate the image type
whether it is an healthy eye or glaucoma- infected eye.
Figure 2: CNN Architecture
3. RESULTS AND DISCUSSION
There are many computer-based techniques are present to identify Glaucoma disease, but these systems have advantages as
well as disadvantages. We have developed a system having more advantages and fewer drawbacks by using a smarter
algorithm. The aim of our project is to improve the performance and accuracy of the system by using a different algorithm.
To develop the proposed system, we have used the Convolutional Neural Network technique which is considered as the best
technique to classify images and detect whether the eye is infected by Glaucoma disease or not.
The process for detecting Glaucoma disease starts by selecting the eye image and then eye image is given as input to the

system and then after processing image on an algorithm we get result whether the eye is infected by Glaucoma or not. The
proposed system can be used by the doctor, lab assistant or any other person who can handle the computerized system
because this system is very much user- friendly.
We have used 100 images for training the algorithm and 100 images for testing. Out of 100 images taken for testing, we have 50
images with a normal eye and 50 images that are infected by Glaucoma.
After testing all images we obtained result (Confusion Matrix) as given below:
Predicted as
Normal
Image
Predicted as
Infected by
Glaucoma
Total
Normal Images TN=48 FP= 2 50
Glaucoma Images FN=1 TP=49 50
49 51 100
Figure 3: Confusion Matrix
TN=True Negative FP=False Positive FN=False Negative TP=True Positive
Precision= TP/ (TP+FP) = 49/ (49+2) = 0.9607
Recall=TP/ (TP+FN) = 49/ (49+1) = 0.98
Accuracy=97%
3.1 IMAGE WITH GLAUCOMA
Figure 4: Output image of Glaucoma detected eye
Figure 5: Detection of Glaucoma
3.2 IMAGE WITHOUT GLAUCOMA
Figure 6: Output image of Healthy eye

Figure 7: Absence of Glaucoma
4. CONCLUSION
In this paper, we have designed and implemented the project using Convolutional Neural Network. Our system will provide a
better solution for detecting Glaucoma in the earlier phase in less time which will save the vision of many people. To implement
this project we have used the Region of Interest (ROI) to take the only region of the image in which Glaucoma can be detected,
also we have used Gaussian blur to remove noise from image and then preprocessed image is given to CNN. Our system gave
0.9607 and 0.98 value of precision and recall respectively. In future, we will try to increase Accuracy of our project by using
different technologies.
REFERENCES
[1]“Detection of Glaucoma Using Retinal Fundus Images” Hafsah Ahmad, Abubakar Yamin, Aqsa Shakeel, Syed Omer Gillani,
Umer Ansari(IEEE 2014)
[2]“Automatic Glaucoma Detection by Using Funduscopic Images”, Atheesan S., Yashothara S.(IEEE 2016)
[3]”Automated Detection of Suspected Glaucoma in Digital Fundus Images”, Namita Sengar, Malay Kishore Dutta, Radim
Burget, Martin Ranjoha(IEEE 2017)
[4] Budai A, Bock R, Maier A, Hornegger J, Michelson G. Robust vessel segmentation in fundus images. Int J Biomed Imag.
2013.
[5] Srivastava N, Hinton G, Krizhevsky A, Sutskever I, Salakhutdinov R. Dropout: a simple way to prevent neural networks
from overfitting.
[6]Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition. 2014. ArXiv e-prints
arxiv:abs/1409.1556.
[7]Carneiro G, NascimentoJ, Bradley AP. In : Navab N, Hornegger J, Wells WM, Frangi AF, eds. Unregistered multiview
mammogram analysis with pre-trained deep learning models.
[8]World Health Organization. Bulletin of the World Health Organization, Volume 82(11). 2004.
[9] Bourne RRA. Worldwide glaucoma through the looking glass. Br J Ophthalmol.2006.
[10]Bock R, Meier J, Nyúl LG, Hornegger J, Michelson G. Glaucoma risk index: automated glaucoma detection from color
fundus images. Med Image Anal. 2010.
[11] Sivaswamy J, Krishnadas SR, Joshi GD, Jain M, Ujjwal A, ST Drishti. Retinal image dataset for optic nerve head (ONH)
segmentation.In:2014IEEE 11th international symposium on biomedical imaging (ISBI), 2014
[12] Chen X, Xu Y, Wong DWK, Wong TY, Liu J. Glaucoma detection based on deep convolutional neural network. In: 2015
37th annual international conference of the IEEE engineering in medicine and biology society (EMBC). 2015. [13] Alghamdi
HS, Tang HL, A.Waheeb S, Peto T. Automatic optic disc abnormality detection in fundus images: a deep learning approach. In:
OMIA3 (MICCAI 2016). 2016

IRJET- Glaucoma Detection using Convolutional Neural Network

Recommended

More Related Content

What's hot (20)

Similar to IRJET- Glaucoma Detection using Convolutional Neural Network (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Glaucoma Detection using Convolutional Neural Network