IOP Conference Series: Materials

Science and Engineering

PAPER • OPEN ACCESS You may also like

- A Data Mining Inspired Methodology
Detection of diabetic retinopathy using deep towards the Identification of Diabetic
Retinopathy
learning methodology KG Pradeep Kumar, K Karunakara, GS
Thyagaraju et al.

- CoT-XNet: contextual transformer with

To cite this article: Gazala Mushtaq and Farheen Siddiqui 2021 IOP Conf. Ser.: Mater. Sci. Eng. 1070 Xception network for diabetic retinopathy
012049 grading
Shuiqing Zhao, Yanan Wu, Mengmeng
Tong et al.

- An Empirical Exploration of Artificial

Intelligence in Medical Domain for
View the article online for updates and enhancements. Prediction and Analysis of Diabetic
Retinopathy: Review
S V Deshmukh and A Roy

This content was downloaded from IP address 103.100.4.239 on 15/08/2024 at 17:56

ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

Detection of diabetic retinopathy using deep learning

methodology

1 2
*
Gazala Mushtaq and Farheen Siddiqui

Department of Computer Science and Engineering, School of Engineering Sciences and

Technology, Jamia Hamdard, New Delhi-110062, India
1*
[email protected], 2 [email protected]

Abstract. Diabetic retinopathy is a complication of diabetes that targets the eyes by damaging
the retinal blood vessels. Initially it is asymptomatic or causes fluctuating vision problems. As it
becomes severe, it affects both the eyes and eventually causes partial or complete vision loss.
Primarily occurs when the blood sugar level is unmanageable. Therefore, the person with
diabetes mellitus is always at a high risk of acquiring this disease. The early detection can deter
the contingency of complete and permanent blindness. Thus, requires an efficient screening
system. The present work considers a deep learning methodology specifically a Densely
Connected Convolutional Network DenseNet-169, which is applied for the early detection of
diabetic retinopathy. It classifies the fundus images based on its severity levels as No DR, Mild,
Moderate, Severe and Proliferative DR. The datasets that are taken into consideration are
Diabetic Retinopathy Detection 2015 and Aptos 2019 Blindness Detection which are both
obtained from Kaggle. The proposed method is accomplished through various steps: Data
Collection, Preprocessing, Augmentation and modelling. Our proposed model achieved 90% of
accuracy. The Regression model was also employed, manifested up an accuracy of 78%. The
main aim of this work is to develop a robust system for detecting DR automatically.

1. Introduction

Diabetes is one of the most common diseases and its prevalence has increased worldwide. It is
primarily associated with the production of insulin and high blood sugar of the body[1] resulting
in the anomalous metabolic functions and complications like cardiovascular diseases, kidney
failures, neural disorders and diabetic retinopathy (loss of vision), etc. Diabetic retinopathy is
a crucial eye condition which results in loss of vision that cannot be reversed or corrected once
experienced. The people who have a long history of diabetes are more prone to get afflicted
with this disease, no matter whether a person is type1 or type2 diabetic, the probability of the
disease increases as the age increases[2]. According to WHO, DR is an intense eye disease that
requires urgent contemplation at an international level. According to a report, in India there are
about 12,000 ophthalmologists for 60 million diabetics with eye disorder. The main reason for
such an alarming number of patients is the result of the fact that mostly people are oblivious
that they are suffering from this disorder. They also show insensitivity and an incautious attitude
towards this disease. About 18% of people with diabetes are suffering from DR and the
possibility of procurement of DR in a diabetic person is 25 times more than that of a healthier
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

one[3]. The detection of this disorder is difficult to diagnose at an initial stage, owing to the
fact that it is asymptomatic or shows very mild symptoms thus leaves a person in oblivion and
eventually leads to vision impairment. Thus to detect DR at an early stage is pivotal in averting
the complexities of this disorder. The diagnostics of this illness requires the professionals and
specialists with highly effective equipment’s and techniques that foster the advancements in
leveraging the prognosis of this condition.

Figure 1. Normal Retina Figure 2. Diabetic Retinopathy

The image of normal retina and the Retina with Diabetic retinopathy is shown in figure 1 and
figure 2 respectively[4]. Since an unerring automatic detection technique is required to classify
and circumspect the severity level of DR. Mostly research in the area of DR was carried out on
the basis of feature extraction using machine learning approaches, but the problem rose with
the manual feature extraction which prompted researchers towards deep learning. The further
research in medical fields paved path for many computer aided technologies like data mining,
image processing, machine learning and deep learning. However, the Deep Learning has gained
popularity in recent years in various fields like sentiment analysis, hand written recognition,
stock market prediction and medical image analysis, etc. CNN in deep learning tends to provide
constructive results when it come to the job of image classification. The architecture of CNN
with its different layers is given in the figure 3[5].

Figure 3. CNN Architecture

The present research employs deep learning methodology especially CNN variant DenseNet,
which extracts the features automatically, rather than manually for the classification of fundus
(eye) images based on the severity level. A Combination of the data set of “Diabetic
Retinopathy Detection” 2015 from Kaggle and “Aptos 2019 Blindness Detection” from Kaggle
was assembled for this study.

2
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

The paper is organized into various sections as: section 2 presents the recent work in this field.
Section 3 gives the proposed solution for detecting DR. Section 4 will discuss the results of the
research. Section 5 will provide a conclusion based on the research carried out in this study and
finally Section 6 will conclude the paper by stating the limitation and future work for further
study.

2. Related Work

Diabetic Retinopathy is one of the grave concerns that engrossed the whole world. Receiving
the attention from various researchers in order to find the optimal solutions for the early
detection of this disease, consequently leading to the prevention of premature fluctuations in
vision. Many studies have been conducted and still continues in this field with an aim to ease
the lives of both doctors as well as patients. This section provides a review of many research
works in the area of Diabetic retinopathy.

J.calleja.et al [6] in their work used a two staged method for Diabetic retinopathy detection:
LBP (Local Binary Patterns ) for feature extraction and Machine Learning specifically SVM
and Random Forest for classification purpose. The results obtained by the random forest
outperformed the SVM with an accuracy of 97.46%. However, the dataset used in this study
was quite small with 71 images.

Earlier works were based on manual feature extraction for detection of DR using various
computer based systems. U.Acharya.et al[7] used features like blood vessels, microaneurysms,
exudates, and haemorrhages from 331 fundus images using SVM with an accuracy of more
than 85%. K. Anant.et al[8] in their literature used texture and wavelet features for DR detection
by making use of data mining and image processing on a database DIARETDB1 and achieved
97.95% accuracy. M.Gandhi.et al[9] proposed a method for automatic DR detection with SVM
classifier by detecting exudates from fundus images. Some works try to integrate manual
feature extraction with deep learning feature extraction for DR. one of such work include
J.Orlando.et al [10] where CNN with hand crafted feature are used for feature extraction for
detecting red lesion in the retina of an eye.

S.Preetha et al.[11] In their literature predicted various diabetic related diseases using Data
Mining and machine learning methods specifically for heart disease and skin cancer prediction
while considering both advantages and disadvantages.

While many researches or works are there about using machine learning approaches or data
mining approaches, a quite different approach also came into the way of detection of diabetic
retinopathy. S.Sadda et al. [12] make use of quantative approach to identify new parameters for
detecting proliferative diabetic retinopathy. It is based on the hypothesis that location, number
and area of lesions can improve the forecasting process of Retinopathy. The methods used for
this study were Subjects and Imaging Data, Ultrawide Field Image Lesion Segmentation,
Quantitative Lesion Parameters and Statistical Analysis. Comparison of lesions were made on
the basis of Lesion number, Lesion surface area, Lesion distance from the ONH center and
Regression analysis. The work presented by J.Amin et al. [13] provides a review of various
methodologies for diabetic retinopathy by detecting hemorrhages, microaneurysms, exudates
and also blood vessels, and analyzes the various results obtained from theses methodologies
experimentally in order to give indepth insight of ongoing research. The study carried out by
Y.Kumaran and C.Patil [14] focuses on the different types of preprocessing and segmentation

3
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

techniques mostly and gives an in detail procedure for detection of diabetic retinopathy in
human eye consisting of number of systems and classifiers. M.Chetoui et al.[15] Proposes a
diagnostic method for DR using machine learning specifically SVM and Texture features.
Texture features used were LTP (Local Ternary Pattern) and LESH (Local Energy-based Shape
Histogram) that provided better results when compared to Local Binary Pattern (LBP). The
accuracy of 90.4% was obtained by LESH with SVM.

Deep learning is the most popular approach among researchers for detection, prediction,
forecasting and classification task in various fields from few years, in medical field particularly
in diabetic retinopathy it is unveiling many possibilities for the prevention of such a dreadful
disease. I.Sadek et al.[16] in their work automatically detected the diabetic retinopathy using
deep learning approach. They used the four convolutional neural network to classify the diabetic
retinopathy into three classes as Normal, Exudates, Drusen. This method outperforms the Bag
of words approach and achieved an accuracy of 91%-92%. G.Zago et al.[17] in their study
designed a lesion localization model using a deep network specifically convolutional neural
network approach with an aim to address the models complexity so, that the performance can
be improved. Instead of segmentation localization process of regions were used. Two
convolutional networks were used for training purpose on a Standard Diabetic Retinopathy
Database and DIARETDB1 where 94% - 95% of sensitivity was obtained. D.Doshi et al.[18]
in their work used a GPU based convolutional neural network to classify the retinal images
severity level into 5 stages. This approach used 3 models of CNN architecture and an ensemble
model of the three to evaluate the results on kappa metrics. The best result were achieved by
ensembling method with a score of 0.3996. The work of P.Kaur et al.[19] Presents a Neural
network technique for the classification of retinal images using MATLAB. The results obtained
were compared with the machine learning approach like SVM where better results were
achieved. M.Voets et al.[20] in their study used a kaggle dataset EyePACS for detection of
diabetic retinopathy from retinal fundus images. However, this study is the re-implementation
of already existing work but on different data set which provided 95% of AUC. The difference
of AUC between the original and the re-implemented method tend to be very large

3. Proposed Methodology

The main objective of this work is to build a stable and noise compatible system for detection
of diabetic retinopathy. This work employs the deep learning methodology for detecting the
diabetic retinopathy based on severity level (No DR, Mild, Moderate, Severe and Proliferative
DR). Many processes were carried out before feeding the images to the network. We trained
two models in this work: our proposed model and the regression model and then a comparison
was made between the accuracies obtained by the two models. Though our proposed model
performed better than the regression model. The figure 4 shows the proposed methodology.

4
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

Data Collection

Data Preprocessing

Data Augmentation

Modelling

Results
Figure 4 Proposed Methodology

3.1 Data Source: Data used for this study has been taken from Diabetic Retinopathy
Detection 2015[21] and APTOS 2019 blindness detection[22] from kaggle. Both the datasets
contains thousands of retinal images under different conditions. For every subject, two images
of both the eyes are given as left and right. As the images come from different sources like
different cameras, different models, etc. It has an abundance of noise associated with it, which
apparently needs to be removed, thus, requiring a number of preprocessing steps. The diabetic
retinopathy associated with each image has been rated on the scale of 0-4 as:

 0 - No DR

 1 - Mild

 2 – Moderate

 3 - Severe

 4 - Proliferative DR

Figure 5 shows the retinal images with ratings on the basis of severity level from 0-1.

(a) (b) (c) (d) (e)

+

Figure 5 Image samples based on severity from dataset: (a) is level ‘0’, (b) is level‘1’, (c) is
level ‘2’, (d) is level ‘3’, (e) is level ‘4’.

5
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

3.2 Data Preprocessing As the images in the dataset contains a lot of noise, like some images
may be out of focus, some may have a lot of exposure, some may have extra lighting, presence
of the black background, etc. so we need to do preprocessing in order to get them in the standard
format. Following things are carried out in preprocessing step:

 Cutting the black border: The black background of the fundus image does
not add any information to the image and is therefore useless so, the black
background around the images are omitted.

 Remove the black corner: After removing the black border there still exist
some black corners as the fundus image is round in shape. In this step black
corners are removed from the image.

 Resizing image: The images are resized to 256*256 (width*height).

 Applying the Gaussian Blur: Gaussian blur is applied to the images by

specifying the kernel size to 256/6. This method helps in removing the
Gaussian noise.

Figure 6 shows the images obtained after preprocessing was carried out.

Figure 6 Images obtained after preprocessing.

3.3 Data Augmentation: After analyzing the data, we notice that the data is highly unbalanced
among the diabetic retinopathy severity image classes as shown in figure 7, which gave rise to
the propensity of data augmentation.

6
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

Figure 7 highly unbalanced data before Figure 8 Balanced data after data augmentation
data augmentation

Data augmentation is framed by aligning one class to the class with most samples, in order to
balance the data among the diabetic retinopathy severity classes, as shown in figure 8. Images
were mirrored and rotated to augment the dataset, 7000 images were obtained in each class after
augmentation shown in figure 9.

Figure 9 Images Obtained after Data Augmentation

3.4 Modelling: We used a DenseNet-169 (Densely connected convolutional neural network)

and Regression model for training purpose. In DenseNet-169 weights are loaded into the
network without the top or last layer. When modelling the network, initially there is no last
layer. We design this layer by using Global Average Pooling 2D, a Dropout layer set at 0.5 and
an output comprising of five nodes for each class. Global Average Pooling 2D is same as that
of 2D average Pooling in operation, but it considers the entire input block size as pool size. A
Dropout layer address the issue of over-fitting. Adam optimization algorithm is used for
optimizing the weights on training this model. A sequential modelling approach is used for
adding layers and customizing the layers like convolutional, dropout, dense, optimizers, etc.

Convolution layer: It employs several kernels or filters to run across the fundus images and
calculate a dot product. Every kernel or filter in this layer draws various image characteristics

7
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

Pooling layer: It provides an abstract representation of convolved features by reducing the

spatial dimension. It is somewhat similar to convolution layer but it takes the max or min region
depends on the type of pooling from kernel-overlapped input.

Dropout layer: The dropout approach has been used to control neural networks in order to
reduce over-fitting.

Flatten layer: Flattening transforms the data to the next layer in a 1-dimensional series.[23].

The figure 10 [24] shows the deep DenseNet-169 model with three Dense blocks and three
transition layers consisting of pooling and convolution layer

Figure 10 DenseNet-169 model with three dense blocks

Figure 11 summary of the model.

The summary of the model is given in figure 11. To stabilize the results a regression model was
also used, which achieved an accuracy of 78%.

3.5 Implementation: The implementation was executed using python language, where a wide
variety of libraries were employed for processing of images and to get acquainted with the
system for creating convolutional neural network like DenseNet-169. The type of library
utilized for image management (like rotation and resizing) and preprocessing was OpenCV[25].
However, the mathematical functions required for the implementation was performed by
NumPy[26]. TensorFlow[27] and Scikit-learn[28] were also used for efficient management of
deep learning models and for defining the model. The implementation of model makes use of
GPU enable device for easier and faster processing.

8
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

4. Results and Discussion

We trained our proposed model using DenseNet-169 on a combination of dataset from Diabetic
Retinopathy Detection 2015[21] and APTOS 2019 blindness detection[22] from kaggle. There
was a lot of noise associated with the images provided by the dataset therefore, preprocessing
was needed. For preprocessing, we first removed the black border of the images in order to
focus more on the fundus image only, black corners of images was also removed, then the
images were resized to a standard format of 256*256 of width and height. At last a Gaussian
blur was applied to the images in order to remove the Gaussian noise. After preprocessing we
analyze that the data is highly unbalanced among the severity classes, majority of data belonged
to the class ‘0’i.e. No DR. in order to address this issue, we used data augmentation, which
gives us 7000 images from each severity class and made the data balanced. After preprocessing
and augmentation of images, data was finally fed to the DenseNet-169 for training the model.
After evaluating our model the training accuracy of 0.953 was obtained, while as validation
accuracy of 0.9034 was achieved. We also calculated the Cohen Kappa score which comes out
to be 0.804. We also applied a regression model to our dataset and compute its validation
saccuracy which is 0.789. Our proposed model outperforms the regression model. The results
of our model are summarized in table 1.

Table 1 Result of the Proposed Model

Training Accuracy Validation Cohen kappa score

Accuracy

95% 90% 80%

The figure 12 shows the comparison between the accuracies obtained by the proposed model
and regression model.

Accuracy
92%
90%
88%
86%
84%
82%
80%
78%
76%
74%
72%
Proposed Model Regression Model

Accuracy

Figure 12 Accuracy Obtained By Proposed Model and Regression Model

9
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

Besides Regression model, the proposed model was compared to a number of machine learning
classifiers like Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Decision Tree
(DT). The results are summarized in the table 2, where accuracies of the different classifiers are
given.

Table 2 Results Obtained By Various Classifiers

CLASSIFIER DATASET ACCURACY DR CLASSES

SVM[29] Messidor, 85.6% Normal, Non PDR,

Diabeticret DB1. PDR.

DT[30] Messidor 85.1% Normal, Mild,

Moderate, Severe.

KNN[29] Messidor, 55.1% Normal, Non PDR,

Diabeticret DB1 PDR.

Regression Diabetic 78% No DR, Mild,

Retinopathy Moderate, Severe
Detection 2015 and Proliferative
&APTOS 2019 DR.
from kaggle.

Proposed Model Diabetic 90% No DR, Mild,

Retinopathy Moderate, Severe
Detection 2015 and Proliferative
&APTOS 2019 DR.
from kaggle.

The proposed model achieves the highest accuracy of 90%, followed by SVM with an accuracy
of 85.6%, Decision Tree with 85.1%, Regression with 78%, and KNN with the least accuracy
of 55.17%.

Figure 13 shows comparison between the classifiers in which the proposed model shows the
highest accuracy among other classifiers.

10
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

Accuracy
100.0%
90.0%
80.0%
70.0%
60.0%
50.0%
40.0%
30.0%
20.0%
10.0%
0.0%
SVM DT KNN Regression Proposed Model

Accuracy

Figure 13 Accuracy Obtained By Various Classifiers.

5. Conclusion

Traditional method for detection of DR is prolonged, challenging and costly, thus many
researches were brought up to automate the detection process by using machine learning and
deep learning approaches. In this work, we presented a comprehensive study of various
methodologies for detecting diabetic retinopathy automatically and attempted to propose our
own deep learning approach for the early diagnosis of retinopathy by using a DenseNet-
169(which is a new CNN architecture, having many deep layers). Two datasets: ‘Diabetic
Retinopathy Detection 2015’and ‘APTOS 2019 blindness detection’ from kaggle were used
together for this study. A lot of preprocessing and augmentation was done to standardize the
data in a desired format and to remove the unwanted noise. Beside DenseNet-169 classifier, we
also used a regression model to draw the comparison between the results. Moreover, machine
learning classifiers like SVM, DT and KNN were compared with the proposed system. Where
the best accuracy among all was obtained by the proposed model and it also classifies the
images into more no of classes. Our proposed model performed better than the regression model
by achieving the accuracy of 90% however, 78% accuracy was yielded by the regression model.

6. Limitation and Future Scope

As there are a number of images taken under different conditions, needs to undergo a lot of
preprocessing and augmentation, some features of image might be missed out, so such
techniques should be used that not only preserve all the tiny important features but at the same
time is able to do a successful pre-processing. Moreover multiple images should be provided
for every patient which would in turn increase the possibility of classifying the images correctly
as more information can be gathered rather than only two images per person. The possibility
of tweaking hyper-parameters is constantly growing with the emergence of new neural
networks through better pooling methods. Such methods can be considered for future work to
uncover the possibilities of increasing performance in this area. Furthermore, using different
networks for training the model by the process of ensemble can also lead towards the better

11
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

results. As different model have their own advantages in terms of performance, if tied together,
can help in improving overall productivity of a system rather than an individual model. We
have used two datasets in our study, using more no of datasets or a combination of various
datasets may improve the generalizability. The deployment of such systems can be done by
using the MobileNet, which is a convolutional neural network for developing mobile
applications. The web applications can be developed that can work for windows, Linux and
Android operating systems as a diabetic retinopathy diagnostic tool.

References
[1] https://www.medicalnewstoday.com/articles/323627.
[2] https://www.mayoclinic.org/diseases-conditions/diabetic-retinopathy/symptoms-
causes/syc-20371611.
[3] S. K. Pandey and V. Sharma, “World diabetes day 2018: Battling the Emerging Epidemic
of Diabetic Retinopathy,” Indian J Ophthalmol.
[4] https://www.health.harvard.edu/a_to_z/retinopathy-a-to-z
[5] https://missinglink.ai/guides/convolutional-neural-networks/convolutional-neural-
networks-architecture-forging-pathways-future/
[6] J. De Calleja, L. Tecuapetla, and M. A. Medina, “LBP and Machine Learning for Diabetic
Retinopathy Detection,” pp. 110–117, 2014.
[7] U. R. Acharya, C. M. Lim, E. Y. K. Ng, C. Chee, and T. Tamura, “Computer-based
detection of diabetes retinopathy stages using digital fundus images,” Proc. Inst. Mech.
Eng. Part H J. Eng. Med., vol. 223, no. 5, pp. 545–553, 2009, doi:
10.1243/09544119JEIM486.
[8] K. A. Anant, T. Ghorpade, and V. Jethani, “Diabetic retinopathy detection through image
mining for type 2 diabetes,” in 2017 International Conference on Computer
Communication and Informatics, ICCCI 2017, 2017, doi: 10.1109/ICCCI.2017.8117738.
[9] M. Gandhi and R. Dhanasekaran, “Diagnosis of diabetic retinopathy using morphological
process and SVM classifier,” Int. Conf. Commun. Signal Process. ICCSP 2013 - Proc.,
pp. 873–877, 2013, doi: 10.1109/iccsp.2013.6577181.
[10] J. I. Orlando, E. Prokofyeva, M. del Fresno, and M. B. Blaschko, “An ensemble deep
learning based approach for red lesion detection in fundus images,” Comput. Methods
Programs Biomed., vol. 153, pp. 115–127, 2018, doi: 10.1016/j.cmpb.2017.10.017.
[11] S. Preetha, N. Chandan, K. Darshan N, and B. Gowrav P, “Diabetes Disease Prediction
Using Machine Learning,” Int. J. Recent trends Eng. Res., vol. 6, no. 5, 2020, doi:
10.23883/IJRTER.2020.6029.65Q5H.
[12] S. R. Sadda et al., “Quantitative assessment of the severity of diabetic retinopathy,” Am.
J. Ophthalmol., 2020, doi: 10.1016/j.ajo.2020.05.021.
[13] J. Amin, M. Sharif, and M. Yasmin, “A Review on Recent Developments for Detection
of Diabetic Retinopathy,” Scientifica, vol. 2016. 2016, doi: 10.1155/2016/6838976.
[14] Y. Kumaran and C. M. Patil, “A brief review of the detection of diabetic retinopathy in
human eyes using pre-processing & segmentation techniques,” International Journal of
Recent Technology and Engineering, vol. 7, no. 4. pp. 310–320, 2018.
[15] M. Chetoui, M. A. Akhloufi, and M. Kardouchi, “Diabetic Retinopathy Detection Using
Machine Learning and Texture Features,” in Canadian Conference on Electrical and
Computer Engineering, 2018, vol. 2018-May, doi: 10.1109/CCECE.2018.8447809.
[16] I. Sadek, M. Elawady, and A. E. R. Shabayek, “Automatic Classification of Bright Retinal
Lesions via Deep Network Features,” pp. 1–20, 2017.

12
ICRIET 2020 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 1070 (2021) 012049 doi:10.1088/1757-899X/1070/1/012049

[17] G. T. Zago, R. V. Andreão, B. Dorizzi, and E. O. Teatini Salles, “Diabetic retinopathy

detection using red lesion localization and convolutional neural networks,” Comput. Biol.
Med., vol. 116, p. 103537, 2020, doi: 10.1016/j.compbiomed.2019.103537.
[18] D. Doshi, A. Shenoy, D. Sidhpura, and P. Gharpure, “Diabetic retinopathy detection using
deep convolutional neural networks,” in International Conference on Computing,
Analytics and Security Trends, CAST 2016, 2017, pp. 261–266, doi:
10.1109/CAST.2016.7914977.
[19] P. Kaur, S. Chatterjee, and D. Singh, “Neural network technique for diabetic retinopathy
detection,” Int. J. Eng. Adv. Technol., vol. 8, no. 6, pp. 440–445, 2019, doi:
10.35940/ijeat.E7835.088619.
[20] M. Voets, K. Møllersen, and L. A. Bongo, “Reproduction study using public data of:
Development and validation of a deep learning algorithm for detection of diabetic
retinopathy in retinal fundus photographs,” PLoS One, vol. 14, no. 6, pp. 1–11, 2019, doi:
10.1371/journal.pone.0217541.
[21] “Diabetic Retinopathy Detection Identify signs of diabetic retinopathy in eye images,”
2015. [Online]. Available: https://www.kaggle.com/c/diabetic-retinopathy-
detection/data.
[22] “APTOS 2019 Blindness Detection Detect diabetic retinopathy to stop blindness before
it’s too late,” 2019. [Online]. Available: https://www.kaggle.com/c/aptos2019-blindness-
detection/data.
[23] https://towardsdatascience.com/the-most-intuitive-and-easiest-guide-for-convolutional-
neural-network-3607be47480.
[24] https://www.kaggle.com/pytorch/densenet169.
[25] OpenCV Library opencv.org. .
[26] http://www.numpy.org/.
[27] https://www.tensorflow.org/.
[28] https://scikit-learn.org/.
[29] S. Gupta, “Diagnosis of Diabetic Retinopathy using Machine Learning,” vol. 3, no. 2,
2015, doi: 10.4172/2311-3278.
[30] E. V Carrera and R. Carrera, “Automated detection of diabetic retinopathy using SVM,”
pp. 6–9, 2017.
[31] “World Health Organization.” [Online]. Available: https://www.who.int/news-room/fact-
sheets/detail/diabetes.

13