Detection of Diabetic Retinopathy in Retinal Image Early Identification using Deep CNN

International Journal of Trend in Scientific Research and Development (IJTSRD)
Volume 7 Issue 2, March-April 2023 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470
@ IJTSRD | Unique Paper ID – IJTSRD55047 | Volume – 7 | Issue – 2 | March-April 2023 Page 557
Detection of Diabetic Retinopathy in Retinal
Image Early Identification using Deep CNN
M. Mukesh Krishnan1
, J. Diofrin2
, M. Vadivel3
1
Assistant Professor, Department of Computer Science and Engineering,
2,3
Student, Department of Computer Science and Engineering,
1, 2,3
Francis Xavier Engineering College, Tirunelveli, Tamil Nadu, India
ABSTRACT
Diabetic Retinopathy, the most common reason of vision loss, is
caused by damage to the small blood vessels in the retina. If
untreated, it may result in varying degrees of vision loss and even
blindness. Since Diabetic Retinopathy is a silent disease that may
cause no symptoms or only mild vision problems, annual eye exams
are crucial for early detection to improve the chances of effective
treatment where fundus cameras are used to capture the retinal
images. However, fundus cameras are too big and heavy to be
transported easily and too costly to be purchased by every health
clinic, so fundus cameras are an inconvenient tool for widespread
screening. Recent technological developments have enabled using
smartphones in designing small-sized, low-power, and affordable
retinal imaging systems to perform Diabetic Retinopathy screening
and automated Diabetic Retinopathy detection using machine
learning and image processing methods. However, Diabetic
Retinopathy detection accuracy depends on the image quality and it
is negatively affected by several factors such as Field of View. Since
smartphone-based retinal imaging systems have much more compact
designs than the traditional fundus cameras, the retina images
captured are likely to be low quality with smaller Field of View As a
result, the smartphone-based retina imaging systems can be used as
an alternative to the direct ophthalmoscope once it tested in the
clinical settings. However, the Field of View of the smartphone-
based retina imaging systems plays an important role in determining
the automatic Diabetic Retinopathy detection accuracy.
How to cite this paper: M. Mukesh
Krishnan | J. Diofrin | M. Vadivel
"Detection of Diabetic Retinopathy in
Retinal Image Early Identification using
Deep CNN"
Published in
International Journal
of Trend in
Scientific Research
and Development
(ijtsrd), ISSN: 2456-
6470, Volume-7 |
Issue-2, April 2023, pp.557-565, URL:
www.ijtsrd.com/papers/ijtsrd55047.pdf
Copyright © 2023 by author (s) and
International Journal of Trend in
Scientific Research and Development
Journal. This is an
Open Access article
distributed under the
terms of the Creative Commons
Attribution License (CC BY 4.0)
(http://creativecommons.org/licenses/by/4.0)
KEYWORDS: Diabetic Retinopathy,
Sensitivity, Accuracy, Specificity,
Deep Convolutional Neural
Networks
I. INTRODUCTION:
Damage to the retina caused by diabetes is referred to
as diabetic nephropathy and is one of the main vision
problems. Vision is severely distorted as a result of
damage to blood vessels within the retina, which
causes fluid leaking and blindness. There are typically
four phases of DR,
The first stage is moderate non - proliferative retina
when only aneurysms can develop. The blood vessels
in the second phase of moderate non-proliferative
blindness become so distorted and swollen that they
can no longer effectively convey blood.
The next phase is Moderate semi-retinal, which
resulted in starved blood flow to the retina due to the
increasing obstruction of more vascular systems
prompting the retina can encourage the development
formation of blood vessels. Proliferative diabetic
retinopathy, the ultimate period, occurs when the
retina's growth characteristics are released into the
vitreous gel that fills the eye and stimulates the
creation ofnew blood vessels along the retina's inner
layer.
There are distinguishing features and qualities
associatedwith each stage. But physicians maybemay
well not take anyof them into account, resulting in an
incorrect diagnosis as a result, the notion of designing
an integrative framework for DR identification is
developed. If DR advances through the worst form, it
might cause total blindness. About 2.6% of allcases of
blindness are due to DR. Patients with diabetes who
have had the condition for a long time are more likely
to have DR. Diabetic individuals should have routine
IJTSRD55047

International Journal of Trend in Scientific Research and Development @ www.ijtsrd.com eISSN: 2456-6470
retinal exams to detect and treat DR early and reduce
the possibilityof impairment DR is distinguished by
the appearance of various types of anomalies in retinal
images. Microaneurysms, hemorrhages and both soft
and tough exudates (EX) are all possible, and other
lesions fall under this category.
Due to the fragility of the vessel walls, the initial
symptom of DR is the appearance of tiny red spots
called microaneurysms (MA) on the retina. Sharp
edges and a dimension of fewer than 125 m
characterize this phenomenon. Figure 1 displays the
classification of MA intosix subtypes by Michael et
al., based on their visibility in AOSLOreflectanceand
traditional fluorescence imaging.
Fig. 1.Normal Vs microaneurysms Image
Patches on the retina that are bigger than 125 m and
havean uneven border are considered hemorrhages
(HM). Figure 2 depicts the two different forms HMs,
which are either referred to as "flame" (superficial
HM) in addition to "blot" (deeper HM).
Fig. 2.Types of Microaneurysms
Hard exudates cause vivid yellow areas on the retina
owing to blood leaking. They have sharp edges and
arepositioned in the outer retina.
The expansion of the neuronal cell results in the
formation of white regions on the retina known as soft
secretions (or cotton wool). The shape might be round
or oval.
MA and HM are red lesions, while brilliant lesions are
soft and difficult exudates (EX). It provides a concise
summary of the five phases Defined as the presence of
such lesions, there are five types of DR: no DR, mild
DR, intermediate DR, extreme DR, and proliferating
DR. Imagesfrom each of the DR phases are shown as
an example in Figure 3.

Automated approaches for DR diagnosis are far more
effective than human diagnostic and may save both
time andmoney. Misdiagnosis is more common with
a manual diagnosis, and it takes more time and effort
to complete.
Fig. 3.Types of Hemorrhage [16]
II. LITERATURE SURVEY
The design and Development of a Novel Blood Vessel
Identification System Regarding Earlier Detection of
Diabetic Retinopathy is the topic of a study presented
by Sumeet Dua. In this article, we provide a unique
approach for evaluating retinal images that relyon the
localized recurrent hierarchical decomposition
attained by using quadtrees and edge thread to find
blood vessels. Blood vessels are depicted by central
and/or cingulate blurring boundaries, which are
distinguished by a noticeable brightness difference.
This feature also significantly aids in reducing false
alarms. The approach reduces the amount of
incorrectlydiscardededgesthataremassivelyessential
whilebeing faster than the present way and using less
space for theedge map [1].
Many individuals across the globe suffer from diabetic
retinopathy, which may lead to blindness. Diabetes
mellitus is a dangerous disorder, and prompt treatment
for patients depends entirely on an accurate early
diagnosis. SeeCollectivelyinterpretable for the article.
Ensemble A deep learning model has been created by
researchers to categorize diabetic retinopathy. To
reduce the bias generated by anyone deep learning
model, the Adaboost approach is also used to
aggregate many deep learning models. The
methodologies proposed in this study are more stable
and ultimately producebetter performance compared
to standalone deep learning models. The results of the
trial supported the advantages of the recommended
approach [2].
Diabetic retinopathy (DR), the far more common eye
ailment in diabetics, is caused by tiny vascular
capillaries in the retina is damaged. Patients with
diabetes for 10 years or more are at higher risk for
both vision loss and diabetes. The main focus of
Diagnostics of Eye Disease is the automatic detection
of diabetic retinopathy. Utilizing Morphology
Processes and SVM Classifier, which assesses the
presence or absence of metabolic byproducts in color
fundus retinal images to assess the severity of lesions.
The focus of this study is exudates since they identify
early indications of diabetic retinopathy. Exudates are
mostly brought on byproteins and lipids that seep into
the retina from the bloodstream as a result of blood
vessel damage. Thus, utilizing a morphological
process, we offer an automated technique in this work
for exudate detection in non-dilated fundus image
retinal photographs. The recommended method
allowed these exudates to be distinguished from the
retinal image and blood vessels. This study not only
defines the disorder but also offers some insight into
how severe the illness is. The information supplied by
the classifier algorithm helps to provide a more
accurate assessment of diabetic retinopathy [3].
Deep learning on fundus images has made automated
monitoring and identification of severe diabetic
retinopathyapractical and affordable alternative (DR).
It has been shown that using an entropy image of the
brightness of funduspictures may help convolutional
neural networks (DCNN) based systems identify
referable DR more accurately. The paper
Identification of Diabetic Retinopathy Using
Bichannel Convolutional Neural Networks suggests
the entropy image computed using the green
component of the retina snapshot. Unsharp masking
(UM) is further employedas a pretreatment before the
calculation of the probability pictures. In addition to
using the properties of both the gray level entropy
images and the greenish component produced
outcomes by UM, DCNN transfer learning will be
employed toenhance the detection skills of attributable
DR. One of the main advantages of utilizing a deep
learning system is the improvement in accuracy in
recognizing and diagnosing diabetic retinopathy. The
suggested method, which makes use of deep learning
to assist eye physicians in making attributable DR
diagnoses, begins with the green channel of the RGB
image and would be helpful to the automated retina
image analysis system [4].
Diabetic retinopathy, a disorder in which the blood
vessels of the retina are damaged, may cause vision
loss. Since there is presently no cure for this disorder,
it is critical to identify retinal damage early by
utilizing a screening technique. According to the

reference [5], segmentation techniques maybe used to
identify blood vessels, exudates, and microaneurysms
in diabetic retinopathy. Performance metrics include
things like sensitivity and specificity. The research on
Diabetic Retinopathy Detection Accuracy produced
encouraging results. Using the One rule and backward
propagation neural networks, diabetic and non-
diabetic digital retinal fundus images are divided into
two categories. This method utilizes a higher number
of attributesavailable compared to previous methods
published in the literature by splitting the retinal
fundus image into four equalsections. The proposed
approach uses the Haar wavelet transform followed by
principal component analysis to effectively select
features. Results on the DIARETDB1 database
demonstrate its suitability, exceeding earlier
approaches [6] with a reliability score of 93.8% for a
deep recurrent neural network and a score of 97.75%
for a specificrule classification.
Amol Prataprao Bhatkar's study focuses on the
identification of diabetic retinopathy in retinal
pictures, and he uses a Multilayer Perceptron Network
(MLPNN) to do this. A technique for automatically
assessing whether a retinal image is normal or
pathological is provided by the MLPNN classifier. A
feature vector with various 09 statistical qualities,
including Entropy, mean, standarddeviation, average,
Euler number, contrast, correlation, energy, and
homogeneity, is produced using the 64-point Discrete
Cosine Transform (DCT). The MLPNN was used to
select the best-quality subset using a training dataset.
This work suggests using the MLPNN classifier
approach to detect diabetes mellitus in retinal images.
Several properties are extracted from retinal images
and used as classifier parameters [7]. The 64-point
DCT and 9 statistical parameters are some of these
characteristics.
Together, references [8] established an innovative
technique for DR early detection via the intricately
describedfractal dimension architecture. For diabetics
who are not proliferating at an early stage, the
evaluation of Macular Optical Coherence imaging
derived from optical coherence tomography
angiography (OCTA) scans is helpful. using the
Support Vector Machine (SVM) algorithm, a machine
learning method under supervision, to automate the
diagnostic procedure and increase its accuracy. The
categorizing approach now has a 99.5% accuracyrate.
In addition to the various stages of diabetic
retinopathy, other retinal diseases that affect the
distribution of arteries or neovascularization may be
categorized using this approach. This work
emphasizes the need for the creation of an automated
system for categorizing NPDR retinal pictures [9].
Additionally, according to this roadmap, analyzing
the skeleton may be a viable method of obtaining
certain fractal characteristics, such as the need to
account for informational and correlational factors
when determining whether a gap exists and when a
branching moment occurs. Additionally, the value for
the derived multifractals has just been released. A
support vector machine gives us a simple
computational recipe that produces accurate early
diabetic retinopathy diagnosis [10].
A suggested AI-based approach for the earlydetection
ofdiabetic retinopathy uses picture data to detect and
classify diabetic retinopathy. This approach uses
machine learning toidentify and categorize diabetes
mellitus in retinal fundus data as Normal, Moderate,
or Proliferative Diabetic Retinopathy (PDR), all of
which are helpful diagnostic and therapeuticindicators
for treating patients [11]. The approach primarily
focuses on binary classification. The recommended
method is intended to categorize fundus images into
two categories: healthy and diseased. The system is
built on the Keras Library and Tensorflow's backend.
After the images were processed, the classification
algorithm's performance significantly improved.
Transfer learning and dynamic learning rate reduction
dependent on validation accuracy areused to increase
the model's accuracy [12]. All Surveywork merits and
demerits are displayed in Table 1.
TABLE I. LITERATURE SURVEY - MERIT AND DEMERITS
Merits Demerits
Without any human oversight, it
automatically recognizes crucialfeatures.
It requires more time to train than atypical DCNN.
It also does well while trying torecognize
images.
More training data are needed thanwith some other
approaches.
Regarding input size, they are not
constrained.
The cost of computation is significant, and
complicated datatraining can be tedious.
less loss of spatial information
It requires a lot of processingpower, especially if
there areseveral images.
Preprocessing an image can improve
visual contrast and density while also
removing noise.
There is no probabilistic basis for the classification
because the support vector classifier places data
points above and below the classifying hyperplane.

III. PROPOSED APPROACH
DCNN there is already an SVM classifier that makes
use of a Machine Learning Algorithm. Normal and
glaucomatous retinal pictures from open-source
databases, together with information from local
hospital databases, havebeen put intothe program, and
the statistical findings have been analyzed.Knowing
how to read and save pictures on computers is crucial.
The retinal picture may be recognized by a computer
down to the pixel level. Figure 4 Class has been
trained on both abnormal and normal visions (example
300 images). Brightness and contrast changes to the
retina images are made before it is displayed [13]. The
term "feature extraction" refers to the process of
extracting meta-level characteristics of a picture,
(Exchangeable image file formatis called metadata the
python code mention to “meta” keyword) such as its
geometry, color palette, and degree of contrast. When
utilized to selectively enhance the diagnosticsystem's
accuracy, improving healthcare data collection and
usage is necessary to increase diagnosis accuracy.
Data must support complicated healthcare choices
and be of high quality, accurate, and individualization
to the patient. it is a powerful tool. The procedure for
creating a training picture isidentical to that used to
create a testing image. Checking the training picture
after the input image. Healthy and aberrant Retinal
pictures may be predicted using providing well in the
machine's next classification DCNN [14]. The primary
goal of the 11 recorded image preprocessing is to
carefully eliminate duplication without negatively
impacting the characteristics that play a crucial part in
the process. Converting a picture to grayscale is the
input.
Fig. 4.Proposed Diagram
A. Feature Extraction
Using GLCM for proposed feature parameters is
shown in Table 2.
TABLE II PROPOSED FEATURE
PARAMETERS
Parameter 40% 60% DISC
Entropy 8.64 7.97 0.080
Prominence 2.34 x 107
1.99 x 107
0.159
Correlation 6.47 7.65 0.166
Variance 856 700 0.200
Shade 117000 888 0.274
Homogeneity 0.107 0.157 0.378
IDM 0.0480 0.0842 0.547
Contrast 968 544 0.560
ASM 2.60 x 10-4
7.20 x 10-4
0.939
B. DCNN Architecture
Multiple architectures, with a 22-layer deep model
being the primary emphasis, were trained and
evaluated to determine the efficacy of DCNNs.
Utilizing a combination of close-to-the-bottom
embeddings and diverse-sized spatial filters, this
highly effective network delivers state-of-the-art
accuracy. Deeper characteristics may be learned by
the network thanks to more convolution layers and
better use of its computational power [15].
In a layered architecture shown figure 5, the first
layer may be responsible for learning about edges,
while the last layer would learn how to understand
hard exudate, a characteristic used for DR
classification. Each convolution layer in the network
is followed by a batch normalization layer to smooth
out any spikes in the data caused by the activation at
the top layer, which specifies complicated conceived
as a way amongst inputs and response variables.
When more feature maps are needed, one batch
normalizingper block is implemented.
The DCNN layers are detailed below shown in
figure 6 isdescribed.
Fig. 5.DCNN Architecture [17]

Fig. 6.DCNN Layers
C. Input Layer
For a DCNN to work, the input picture must already
have some kind of image data. As we saw before, a
three- dimensional matrix is used to represent image
data. Make one long column out of it. If your picture
has the dimensions2828 = 784, you'll need to multiply
it by 1 to get a size suitable for input. Input dimension
equals the number of training instances multiplied by
m. (784, m).
D. Convolution Layer
Two-dimensional convolution is used to process the
inputs. Cross-channel integration of the "dot products"
of weights and inputs. Receptive fields may share the
same filter weights. The quantity of information
includes the same "quality" as the collection of
variables, which is the same as thenumber of channels
in the input volume.
E. Pooling Layer
We get activation maps from convolutional layers,
then apooling layer aggressively downsamples those
activation maps (discarding information in the
process), and finally, we employ smaller filters and
stop pooling altogether.
F. Fully Connected Layer
Think of it as the last step in the learning process,
whereby retrieved visual characteristics are mapped
to the intended outcomes. Typically, they adapt to the
demands of encoding and classifying data. The most
often seen form of output is vectors, which itself is
subsequently fed into Softmax to be used as a
representation of classification confidence. The
neuronal network and brain imaging both play a role
in a fully linked layer. It bridges the gap between
different layers of neurons. Images may be trained
into numerous categories using this method.
G. Softmax Layer
A convolutional neural network's last layer is called
the Softmax or Logistic layer. The FC layer it belongs
to ends there. Binary classification can be handled by
logistics, whereas multi-classification can be handled
by softmax. The labeled brain picture is sent to the
output layer as a single-hotrepresentation. You should
be familiar with DCNN now.
IV. RESULTS AND ANALYSIS
Keras is a Python framework that provides a high-
level API for neural networks that can run atop
Tensor Flow, Scikit-learn, or Theano. Fast
experimentation was a primary design goal
throughout development Figures 7 to 14 all are
current research work outputs.
Fig. 7.Input Image
Fig. 8.Pre-processed Image
Fig. 9.Histogram for Disease Prediction

Fig. 10. Conv2D Parameter
Fig. 11. Epochs
Fig. 12. Model Accuracy
Fig. 13. Model Loss
Fig. 14. Comparative Analysis for Existing Vs
Proposed
To test our suggested model with this new dataset,
we use some algorithms built into the Jupyter
notebook. The data as a whole demonstrate the
proposed model's suitabilityfor DR prediction using
this new dataset shown in Figures 12 and 13.
Sensitivity(SN), specificity(SP), and accuracy(ACC)
were utilized in this investigation, and true positive
(TP) refers to the total number of positively detected
samples in the positive set. The number of
classification- negative samples in the negative set is
referred to as true negative (TP). The number of
recognized positive samplesin the negative group is
known as a false positive (FP). The number of
recognized negative samples in the positive set is
represented by the term false negative (FN). It is
frequently employed to assess the caliber of
classification models. The percentage of samples that
the classifier successfully classifies to the total
number of samples is known as the accuracy.
According to the total number of samples, the
classifier successfully categorized. Sensitivity and
specificity are the two fundamental properties of
medical statistics. The true positive rate is sensitivity,
and the true negative rate is specificity.
V. CONCLUSION AND FUTURE WORK
Using deep learning on fundus photos, this study
effectively diagnoses diabetes, and may one day be
utilized as a means for making such a diagnosis. The
promise of convolutional neural networks (DCNNs) is
that they can understand from raw pixels and so make
use of the massive volumes of pictures that have been
processed for human- interpreted screening. DCNNs
may be able to identify other illnesses thanks to the
large heterogeneityand moderate biasof these models.
If you look at visualization discover that the indicators
used for classification are situated in a section of the
characteristics learned by DCNNs. Data is easily
discernible to the naked eye. Images of the retina in
moderate and severe cases of diabetes include
macroscopiccharacteristics at a scale that is above the
capabilities of existing DCNN designs, both for

training and validation. A larger amount of processed
data may be used to train the model using the system
in the future, yielding more accurateresults. There is
an increasing number of people accompanying type 2
diabetes, raising the risk of bleeding such as
retinopathy. An estimated 35% of the 285 million
individuals worldwide who have diabetes also have
diabetic retinopathy, with another 30% of those cases
(about 3.2 million people) being severe enough to
jeopardize sight.
These days, computer-aided methods for the diagnosis
of diabetic retinopathy rely heavily on image-
processing techniques that make use of deep learning.
Some potential courses of action might improve the
efficiency with which deep learning methods are used.
However, ophthalmologists are required for the
examination and characterization of image data. Most
data analysis work has been carried out viathe use of
neural network convolutional models to compound
growth multi-layer foundations for the diagnostic test
of macular degeneration using cloud-based retinal
fundus images. As a result of developments in diabetic
and blood pressure management, as well as laser
treatment and intraocular medication administration,
the likelihood of visual impairment from macular
degeneration has decreased considerably during the
last three decades. However,diabetes continues to be a
leading factor in visual impairment. This report
reviews the current knowledge on diabetic retinopathy
and discusses potential avenues for further study.
REFERENCES
[1] S. Dua, N. Kandiraju and H. W. Thompson,
"Design and implementation of a unique blood-
vessel detection algorithm towards early
diagnosis of diabetic retinopathy," International
Conference on Information Technology:
Coding and Computing (ITCC'05) - VolumeII,
vol. 1, pp. 26-31, 2005.
[2] Jiang, K. Yang, M. Gao, D. Zhang, H. Ma, and
W. Qian, "An Interpretable Ensemble Deep
Learning Model for Diabetic Retinopathy
Disease Classification," 2019 41st Annual
International Conference of the IEEE
Engineering in Medicine and Biology Society
(EMBC), pp. 2045-2048, 2019.
[3] M. Gandhi and R. Dhanasekaran, "Diagnosis of
diabetic retinopathy using morphological
process and SVM classifier," 2013 International
Conference on Communication and Signal
Processing, pp. 873-877, 2013.
[4] L. Pan and X. Chen, "Retinal OCT Image
Registration: Methods and Applications," in
IEEE Reviews in Biomedical Engineering,
2021.
[5] P. Shu-I, L. Hong-Zin, C. Ke-Hung, T. Ming-
Cheng, C. Jiann-Torng, and L. Gen-Min,
“Detection of Diabetic Retinopathy Using
Bichannel Convolutional Neural Network,”
Journal of Ophthalmology, Hindawi,vol. 2020,
2020.
[6] M. T. Vyshnav, V. Sowmya, E. A.
Gopalakrishnan, S. Variyar V.V., K. Menon
and K. Soman, "Deep Learning Based
Approach for Multiple Myeloma Detection,"
2020 11th International Conference on
Computing, Communication and Networking
Technologies (ICCCNT), pp. 1-7, 2020.
[7] D. K. Prasad, L. Vibha and K. R. Venugopal,
"Early detection of diabetic retinopathy from
digital retinal fundus images," 2015 IEEE
Recent Advances in Intelligent Computational
Systems (RAICS), pp. 240-245, 2015.
[8] A. P. Bhatkar and G. U. Kharat, "Detection of
Diabetic Retinopathyin Retinal Images Using
MLP Classifier," 2015 IEEE International
Symposium on Nanoelectronic and Information
Systems, pp. 331- 335, 2015.
[9] Y. Ma et al., "ROSE: A Retinal OCT-
Angiography Vessel Segmentation Dataset and
New Model," in IEEE Transactions on Medical
Imaging, vol. 40, no. 3, pp. 928-939, 2021.
[10] M. Gandhi and R. Dhanasekaran, "Diagnosis of
diabetic retinopathy using morphological
process and SVM classifier," 2013 International
Conference on Communication and Signal
Processing, pp. 873-877, 2013.
[11] M. Shelar, S. Gaitonde, A. Senthilkumar, M.
Mundra, and A. Sarang, "Detection of Diabetic
Retinopathy and its Classification from the
Fundus Images," 2021 International Conference
on Computer Communication and Informatics
(ICCCI), pp. 1-6, 2021.
[12] K. Simy Mary, D. Sujitha Juliet, and V.
Vinodh, “Brain Tumour Detection by Gamma
DeNoised Wavelet Segmented Entropy
Classifier,” Computers, Materials Continua,
vol. 69, no. 2, 2021.
[13] Y. Huang, W. Xia, Z. Lu, Y. Liu; H. Chen; J.
Zhou, L. Fang and Y. Zhang, "Noise-Powered
Disentangled Representation for Unsupervised
Speckle Reduction of Optical Coherence
Tomography Images," in IEEE Transactions on
Medical Imaging, vol. 40, no. 10, pp. 2600-
2614, 2021.

[14] X. Guo, F. Wang, G. Teodoro, A. B. Farris and
J. Kong, "Liver Steatosis Segmentation With
Deep Learning Methods," 2019 IEEE 16th
International Symposium on Biomedical
Imaging (ISBI 2019), pp. 24-27, 2019.
[15] M. W. L. Moreira, J. J. P. C. Rodrigues, V.
Korotaev, J. Al-Muhtadi and N. Kumar, "A
Comprehensive Review on Smart Decision
Support Systems for Health Care," in IEEE
Systems Journal, vol. 13, no. 3, pp. 3536-3545,
2019.
[16] Qummar, Sehrish, Fiaz Gul Khan, Sajid Shah,
Ahmad Khan, Shahaboddin Shamshirband, Zia
Ur Rehman, Iftikhar Ahmed Khan, and Waqas
Jadoon. "A deep learning ensemble approach
for diabetic retinopathy detection." Ieee Access
7 (2019): 150530-150539.
[17] Afridi, Tariq Habib, Aftab Alam, Muhammad
Numan Khan, Jawad Khan, and Young-Koo
Lee. "A multimodal memes classification: A
survey and open research issues." In The
Proceedings of the Third International
Conference on Smart City Applications, pp.
1451-1466. Springer, Cham, 2021.

Detection of Diabetic Retinopathy in Retinal Image Early Identification using Deep CNN

More Related Content

Similar to Detection of Diabetic Retinopathy in Retinal Image Early Identification using Deep CNN (20)

More from ijtsrd (20)

Recently uploaded (20)

Detection of Diabetic Retinopathy in Retinal Image Early Identification using Deep CNN