This document discusses using deep learning techniques to detect diabetic retinopathy from retinal images. Diabetic retinopathy is a disease affecting the retina caused by complications of diabetes. It can lead to vision loss if not managed properly. The document proposes using methods like image preprocessing, feature extraction, segmentation, and classification with deep learning algorithms to automatically analyze retinal images and detect signs of diabetic retinopathy. Specifically, it involves extracting features from images using techniques like color density histograms, then training a deep learning model on those features to classify images according to the severity of diabetic retinopathy present. The goal is to enable early detection of the disease from retinal images to help prevent vision loss.

1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 03 | Mar-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 2504
Eye Diabetic Retinopathy by Using Deep Learning
Amit Kesar1, Navneet kaur2, Prabhjit singh3
1Student, Dept of Computer Science Engineering, GIMET Amritsar, Punjab, India
2,3Assistant Professor, Dept of Computer Science Engineering, GIMET Amritsar, Punjab, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract -Diabetic retinopathy is an ailment, caused by turn
in the retinal veins. It is a strong sign of early visualimpedance
and in case it isn't managed may tend to complete visual
inadequacy and the vision lost once can't be re-built up
eventually. In this paper various picture getting readysystems
are used to isolate between the normal and the contaminated
picture. The undertaking is made to see where the issue truly
lies so authentic finish of patient ought to be conceivable.
Planning of a photo, optic plate acknowledgment, Blood
vessels extraction, Exudates ID are a bit of the techniques that
are associated here. Distinctive counts are planned to get the
pined for result. A sweeping number of massesareimpactedby
this ailment around the world. In our audit paper to
accomplish improve comes about we utilize profoundlearning
technique. This gives us early discovery of sickness so we can
get free off inside time.
Key Words: Diabetic retinography, exudates, profound
learning
1. INTRODUCTION
Diabetes remains standout among the most broadly
perceived disease businesses of visualimpedanceamongthe
overall public of working social occasions. It causes
waterfall, glaucoma and mischief of the veins inside the eye,
such condition is called "Diabetic Retinopathy'. Diabetic
Retinopathy is an exceptional retinal issue whichcausessign
of diabetes on the retina. Around 210 million people
wherever all through the world have Diabetes Mellitus;
among which 10 - 18% of people are encountering Diabetic
Retinopathy. Gargeya et al. [1] so for the neutralizingactivity
of Diabetic Retinopathy and dynamic vision setback, early
revelation and investigation of Diabetic Retinopathy is
required. Human Eye lookslikea camera. Light goesintothe
eye through cornea and iris and it is locked in to the retina
through point of convergence which is accessible among iris
and retina. Retina interprets the light and transmitted it to
the psyche through optical circle. (Report 2014) Finding of
DR is performed by the evaluation of retinal (fundus)
pictures. Manual surveying of these photos to choose the
reality of DR is genuinely direct and resource asking for
(A.S.Panayides et al. 2016) [4]. It happens when diabetes
hurts the little veins inside the retina, the light tricky tissue
at the back of the eye. This unobtrusive vein will spill blood
and fluid on the retina outlines features, for instance, little
scale aneurysms, hemorrhages, hard exudates, cotton wool
spots or venous circles (Al-ayyoub et al. 2016).
Diabetic retinopathy can be completely assigned non-
proliferative diabetic retinopathy (NPDR) (Figure 1b) and
proliferative diabetic retinopathy (PDR). (Teoh et al.
2013)Contingent upon the proximity of features on the
retina, which is said over, the periods of DR can be
perceived. A run of the mill retina of the eye does not have
any of the above said features. In the NPDR arranges, the
ailment can advance from smooth, direct to genuine stage
with various levels of features said above beside less
improvement of crisp volunteer’s vessels. PDR is the
impelled compose where the fluids sent by the retina for
sustenance trigger the advancement of new enlists vessels.
(Bulsara et al. 2011)They create along the retina and over
the surface of the unmistakable, vitreous gel that fills inside
the eye. If they discharge blood, genuine visionadversityand
even visual lack can come to fruition.
1.1 General step in the analysis of image artefacts are
listed as under:
I. Pre-processing
This stage used as a piece of demand to channel the
uproarious bit of the picture and update the picture set
showed for examination. Hullabaloo et al. dealing with
framework that could be combined fuses center isolating,
Gaussian smoothening, adaptable centre channel and not
long[2] after the pre-preparing stage, feature extractionpart
is performed.
II. Feature extraction
This system used as a piece of demand to remove the
fundamental highlights good and gone. These highlights are
used to isolate significant information about the malady
present inside the image. [3]Highlight extricated could
incorporate Mean, Median, Mode, Kurtosis, Standard
Deviation, mean deviation and so forth.
III. Segmentation
Features are expelled from update picture are broke down.
Fundamental bit of the picture is evacuated and senseless
parts are kept away from the picture. The crucial parts are
addressed with a white area and pointless parts are
addressed with dull section.[4] After the picture is isolated
the features are emptied once more. These features are
considered against the readiness set features remembering
the ultimate objective to recognize the damage accepting
any.
IV. Classification
The features(Mean, Median, Mode, Kurtosis, Standard
Deviation, mean deviation and so on.) values so extricated
are analysed against the infection qualities. These attributes

2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 03 | Mar-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 2505
are inspected for fitting in classes of ailment. [5]On the off
chance that ailment fallsinto the classificationofanyailment
then illness is anticipated. For arrangement,calculationslike
K-implies, Decision Tree, SVM and so forth can be utilized.
With a specific end goal to handle the expansive datasets
extraordinary branch of machine learning known as Deep
learning can be utilized. The basic normal for profound
learning is the treatment of extensive dataset. Issue with
profound learning is the treatment of little dataset.
[6]However size of dataset exhibitedforassessmentcouldbe
sufficiently huge to be handled through machinelearning.So
this field of machine learning gives helpful system to
breaking down Diabetic Retinopathy. [7]
2. BACKGROUND ANALYSIS
The current examinations and different analysts have
explored the area of automated conclusion of Diabetic
Retinopathy. This survey, hence, is an endeavour to
fundamentally investigate the writing in the region of
machine learning and profound learning strategies utilized
for DR recognition finishing the procedure with inference of
a correlation between the two. (Gulshan et al, 2016,
Development and Validation of a Deep Learning Algorithm
for Detection of Diabetic Retinopathy in Retinal Fundus
Photographs) The given paper connected profound figuring
out how to develop a calculation for mechanizedrecognition
of diabetic retinopathy alongside diabetic macular edema in
retinal fundus photos. Profound convolutionalneuralsystem
was utilized for image arrangement prepared utilizing a
dataset of 128175 retinal images reviewed 3 to 7 times for
diabetic retinopathy and diabetic macular edema by a board
of 54 US authorized ophthalmologists and ophthalmology
senior inhabitants amongst May and December 2015.The
calculation was likewise approved in January and February
2016 utilizing 2 isolate datasets. The calculation so created
had high affectability and specificity for distinguishing
referable diabetic retinopathy. Additionally inquire about is
important to decide the practicality of applying this setup in
the clinical condition. [13] (Jen Hong Tan et al, 2017,
Automated Segmentation of Exudates, Hemorrhages,
Microaneurysms utilizing Single Convolutional Neural
Network) The paper proposes to utilize a 10-layer
convolutional neural system to naturally simultaneously
fragment and separate exudates, hemorrhages and small
scale aneurysms. Information images were standardized
before division. The net is prepared in two phases to
enhance execution. 30,275,903 compelling focuses in the
CLEOPATRA database accomplishing palatable outcomesin
spite of the fact that a drop in affectability is watched for
hemorrhages and smaller scaleaneurysms.Thisexamination
demonstrates that it is conceivable to get a solitary
convolutional neural system to portion such neurotic
highlights on an extensive variety of fundus images with
sensible precision.
[26] (Doshi et al, 2016, Diabetic Retinopathy Detection
utilizing Deep Convolutional Neural Networks) this paper
goes for programmed determination of DR into various
stages utilizing profound learning. The plan and usage of
GPU quickened profound convolutional neural systems to
consequently analyse has been displayed subsequently
orderinghigh-determination retinal imagesinto 5 phasesof
the sickness in view of seriousness. Three noteworthy CNN
models were composed their designs built and the relating
quadratic kappa wasfound. The singlemodelaccomplisheda
precision of 0.386 on a quadratic weighted kappametricand
outfit of three such comparative models brought about a
score of 0.3996.[25] (Abbas et al, 2017, Automatic
acknowledgment of seriousness level for finding of diabetic
retinopathy utilizing profound visual highlights) In this
article, a novel programmedacknowledgmentframeworkfor
ordering diabetic retinopathy (SLDR) in five seriousness
levels is created through learning of profound visual
highlights (DVFs).However, no pre-or post-processing steps
were performed. Extraction of DVF highlightswasdonefrom
each image by utilizing shading thick in scale-invariant and
slope area introduction histogram strategies. Learning
included a semi-administered multilayer profound learning
calculation alongside another compactedlayerandadjusting
steps. This SLDR framework was assessed and contrasted
and cutting edge procedures utilizing the measures of
affectability (SE), specificity (SP) and zone under the getting
working bends (AUC). Around 750 fundus images were
broke down and exceptionally fulfilling outcomes were
gotten which exhibited that theproposedSLDRframeworkis
suitable for early discovery of DR and give a powerful
treatment to forecast kind of diabetes.(Gargeya et al, 2016,
Automated Identification of DiabeticRetinopathyUsingDeep
Learning) This paper displaysthe improvementtookafterby
an assessment of an information driven profound learning
calculation as a demonstrative device for mechanized DR
discovery. The calculation prepared shading fundus images
furthermore grouping them as sound (no retinopathy) or
having DR, recognizable proof of cases significant for
therapeutic referral is additionally done. An aggregate of 75
137 freely accessible fundus images from diabetic patients
were utilized to prepare and test this model toseparatesolid
fundi from those with DR. A board of retinal experts decided
the ground truth for the given informational index before
experimentation. The model was likewise tried utilizing the
general population MESSIDOR 2 and E-Ophthadatabasesfor
approval reason. Representation of the data learned was
finished utilizing a consequently created variation from the
norm warm guide. The model accomplished satisfactoryand
precise outcomes.
[24] (Grinsven, 2016, Fast convolutional neural system
preparing utilizing specific information examining:
Application to drain identification in shading fundus
images).The proposed strategy givesachangeandaccelerate
of the CNNprepared for medicinal image examination
errands by progressively choosing misclassified negative
examples. Heuristic inspecting of preparing tests is done in
light of order by the present status of the CNN. Weights are
then doled out to the preparation tests. A correlation has
likewise been performed between the two i.e., CNN with

3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 03 | Mar-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 2506
(SeS) and without (NSeS) the specific inspecting strategy.
Concentrate is on the location of hemorrhages in shading
fundus images. Execution was enhanced and agreeable and
an equivalent zone under bend trademark was achieved on
two informational collections. Be that asit may, the SeS CNN
measurably beat the NSeS CNN on an autonomous test set.
[23](Quellec et al, 2017, Deep Image Mining for Diabetic
Retinopathy Screening) Here, a speculation of the back
propagation strategy is proposed so asto prepare ConvNets
that deliver fantastic heatmaps. The proposed arrangement
is connected to diabetic retinopathy (DR) screening in a
dataset of right around 90,000 fundusphotosfrom the2015
Kaggle Diabetic Retinopathy rivalry and a private dataset of
just about 110,000 photos(e-ophtha). For the assignmentof
recognizing referable DR, great location execution was
accomplished. Execution was additionally assessed at the
image level and at the injury level in the DiaretDB1 dataset,
where the proposed finder prepared to recognize referable
DR outflanks late calculations prepared to distinguish those
sores especially, with pixel-level supervision. At the sore
level, the proposed finder beatsheatmap agecalculationsfor
ConvNets.
[22] (Pratt et al, 2016, Convolutional Neural Networks for
Diabetic Retinopathy) In this paper, CNN approach is
proposed to diagnosing DR from computerized fundus
images alongside precise arrangement of its seriousness. A
system with CNN engineering and information growth is
created which recognizes highlights, for example,
miniaturized scale aneurysms, exudate and hemorrhageson
the retina engaged with the order errand and as a resultgive
an analysis naturally without client input. To prepare this
system, a top of the line illustrations processor unit (GPU) is
utilized on the freely accessible Kaggle dataset with 80,000
images which showed amazing outcomes, especially for an
abnormal state grouping assignment. Additionally, 5000
images are utilized for approval which likewise exhibited
appropriate results.
[6] (Colas E, 2016, Deep learning approach for diabetic
retinopathy screening) The proposed technique called
Dreamup Vision utilizes best in class innovation in light of
profound learning. The given calculation was prepared on
more than 70,000 named retinal images. Images were
reviewed by ophthalmologists as takes after: 0 (no
retinopathy), 1 (gentle non proliferative DR), 2 (direct non
proliferative DR), 3 (extreme non proliferative DR) and 4
(proliferative retinopathy). Both left and right eye portrayal
was utilized. Evaluating is improved the situation each eye
image independently. The calculation performedspeedyand
solid discovery of abnormalities in retinal images, analysed
their phase of diabetic retinopathy and gave the area of the
oddities identified in the photos. A patient wasconsideredas
referable if the DR organize is in the vicinity of 2 and 4 and
non-referable something else. The model is assessed on
more than 10,000 fundus images from 5,000 patients taken
from the Kaggle DR Detection Challenge dataset, gave by
California Healthcare Foundation.
[8] (Roy et al, 2017,anovel crossover approach for
seriousness evaluation of diabetic retinopathy in shading
fundus images) In this paper, proposition is to consolidate
CNNs with word reference based methodologies, which
coordinates pathology particular image portrayal into the
learning system for an enhanced order of DR seriousness.
Both particular and generative pathology histograms were
built and joined with include portrayals separated from
completely associated CNN layers. The outcomes
demonstrated that the proposed strategy indicated
perceptible change in quadratic kappa score. (Paul et al,
2016, Heterogeneous Modular Deep Neural Network for
Diabetic Retinopathy Detection) This paper proposes
heterogeneous secluded profound neural system (DNN) to
distinguish diabetic retinopathy alongside the five sorts of
variations from the norm. The particular approach givesthe
preferred standpoint to remove class particular highlights
for the classifier, henceforth, beating the traditional
convolutional neural systems. Additionally, the
heterogeneous idea of measured DNN guarantees economy
of scale in the general engineering and furthermore
accommodatesextraction of area particularhighlightswhich
additionally add to higher precision of discovery. The
benchmark dataset was utilized for performing broad
reproduction studies and results demonstrated that the
proposed approach performs better or at standard with
other standard methodologies.
(Takahashi et al, 2017, Applyingcounterfeitconsciousnessto
ailment arranging: Deep learning for enhancedorganizingof
diabetic retinopathy) The review consider examined 9,939
back shaft photos of 2,740 patients with diabetes. Non-
mydriatic 45° field shading fundusphotosweretakenoffour
fields in each eye every year at Jichi Medical University
between May 2011 and June 2015. An altered completely
arbitrarily instated GoogleNet profound learning neural
system wasprepared on 95%of the photosutilizing manual
adjusted Davis reviewing of three extra contiguous photos.
The reviewing of 4,709 of the 9,939 back shaftfundusphotos
was finished utilizing genuine visualizations. Furthermore,
95% of the photos were learned by the adjusted GoogleNet.
Principle result measures were commonness and
predisposition balanced Fleiss' kappa (PABAK) of AI
organizing of the rest of the 5% of the photos. (Eltanboly
et al, 2016, A PC supported symptomatic framework for
recognizing diabetic retinopathy in optical soundness
tomography images) The proposed PC supported
symptomatic (CAD) framework for recognizing diabetic
retinopathy (DR) contrasts from above expressed strategies
as it consolidates novel and productive division and
arrangement methods on optical soundness tomography
(OCT) images of the retina. In the wake of fragmenting 12
unmistakable retinal layers, three quantitative highlights
(pixel-wise ebb and flow, reflectivity and thickness) are
removed each finished a claim layer indicatingfactuallyhuge
contrasts of qualities amongst ordinary and diabetic
subjects. Combined conveyance elementsof thesehighlights
are examined by a two-organize trainable profound
combination grouping system (DFCN) with heaps of non-

4. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 03 | Mar-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 2507
pessimism obliged auto encoders (SNCAE) to decide if the
subject has DR or not. Both the quantitative and visual
evaluations affirmed the high exactness of the proposed PC
helped symptomatic framework for early DR recognition
utilizing the OCT retinal images. [12] (Prentasic et al, 2016,
Detection of exudates in fundus photos utilizing profound
neural systems and anatomical point of interest
identification combination) Profound convolutional neural
systems have been utilized for exudate discovery.Keepingin
mind the end goal to incorporate abnormal state anatomical
information about potential exudate areas, yield of the
convolutional neural system is joined with the yield of the
optic plate discovery and vessel identification methods. In
the approval step utilizing a physically sectioned image
database, a most extreme F1 measure of 0.78 was acquired.
This, in this way, has been turned out to be an imperative
advance in making computerized screening programs for
early discovery of diabetic retinopathy[21]. (Prentasic et al,
2015, Detection of Exudatesin FundusPhotographsutilizing
Convolutional Neural Networks) In this paper it is
demonstrated that profound convolutional neural systems
can be adequately utilized as a part of request to section
exudates in shading fundusphotos. Be that asit may, thelast
division result would positively enhance by upgrading the
system by utilizing all the accessible channels and including
some pre-processing and post processing steps since
gathering recognized pixels in bunches and including some
abnormal state highlights.
[15] (Shan et al, 2016, A profound learning strategy for
micro aneurysm location in fundus images) In this paper, a
two layered stacked space auto-encoder structure is given
for programmed MA location on fundus images. The SSAE
model can catch abnormal state highlights from pixel level
contribution to an unsupervised learning strategy. These
abnormal state highlights gives proficiency to the SMC
classifier and empower it to distinguish MA sores from
fundus images with entangled background. The execution
when adjusting were analysed and it was seen that the
tweaking ended up being useful in boosting the execution of
the fix order. [16] (Holly H. Vo et al, 2016, New Deep Neural
Nets for Fine-Grained Diabetic Retinopathy Recognition on
Hybrid Colour Space) In this paper, the part of numerous
filter sizes in learning fine-grained discriminanthighlightsis
stressed henceforth proposing two profound convolutional
neural systems - Combined Kernels with Multiple Losses
Network (CKML Net) and VGGNet with Extra Kernel(VNXK),
which are a change upon GoogleNet and VGGNet insettingof
DR undertakings. Additionally, half breedshadingspace,LGI,
for DR acknowledgment is proposed and ultimately,
exchange learning is connected to take care of the issue of
imbalanced dataset. The viability of proposed framework is
assessed utilizing two thousand test retina datasets:
EyePACS and Messidor.
[18] (Costa et al, 2017, Convolutional Bag of Words for
Diabetic Retinopathy Detection from Eye Fundus
Images).This paper depicts a system for Diabetic
Retinopathy discovery from eye fundus images utilizing a
speculation of the Bag-of-Visual-Words (BOVW) strategy.
The plan of the BVW includestwo neural systemsthatcanbe
prepared mutually. Not at all like the BOVW, the proposed
show can figure out how to perform include extraction,
highlight encoding and characterization through the
grouping mistake. The model accomplishes an enhanced
Area under the Curve (AUC) on the DR2 and Messidor
datasetswhen contrasted withthestandardBOVWapproach
[19].
3. CONCLUSION AND FUTURE SCOPE
Deep Learning instrument can deal with substantial dataset.
However slight change in the present dataset could prompt
radical change in result because of vagueness issue show
inside existing writing. Keeping in mind the end goal to
handle the issue, stable information introduction by taking
care of commotion inside the image is considered inthought
about approach. The choice tree order is mind boggling and
could lead to model. Overall classification accuracy could
improve by the application SVM.
REFERENCES
1. Ophthalmology, pp. 1–8, 2017.
2. N. Singla, “A Comparative Study of Noising And
Denoising Technique In Image Processing,” vol. 4, no. 3,
pp. 38–42, 2016.
3. M. Raghav and S. Raheja, “IMAGE DENOISING
TECHNIQUES : LITERATURE REVIEW,” vol. 3, no. 5,
2014.
4. A. S. Panayides, C. S. Pattichis, and M. S. Pattichis, “The
Promise of Big Data Technologies and Challenges for
Image and Video Analytics in Healthcare,” pp. 1278–
1282, 2016.
5. S. Shrestha, “Image Denoising Using New Adaptive
Based Median Filter,” Signal Image Process., vol. 5, no.4,
pp. 1–13, 2014.
6. P. D. Kaur and I. Chana, “Cloud based intelligent system
for delivering health care asa service,”Comput.Methods
Programs Biomed., vol. 113, no. 1, pp. 346–359, 2013.
7. M. S. Haleem, L. R. Gargeya and T. Leng, “Automated
Identi fi cation of Diabetic Retinopathy Using Deep
Learning,” Han, J. Van Hemert, B. Li, and A. Fleming,
“Retinal Area Detector from Scanning Laser
Ophthalmoscope ( SLO ) Images for Diagnosing Retinal
Diseases,” vol. 2194, no. MARCH, 2014.
8. “1-s2.0-S2001037016300733-main(1).pdf.” .
9. Tan, J. H., Fujita, H., Sivaprasad, S. Bhandary, S. V., Rao,A.
K., Chua, K. C., & Acharya, U. R. (2017). Automated
segmentation of exudates, haemorrhages, micro

5. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 03 | Mar-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 2508
aneurysms using single convolutional neural network.
Information Sciences, 420, 66-76. DOI:
10.1016/j.ins.2017.08.050
10. Van Grinsven, M. J., van Ginneken, B., Hoyng, C. B.,
Theelen, T., & Sánchez, C. I. (2016). Fast
convolutional neural network training using selective
data sampling: Application to hemorrhage detection in
color fundus images. IEEE transactions on medical
imaging, 35(5), 1273-1284. DOI
10.1109/TMI.2016.252668
11. Abbas, Q., Fondon, I., Sarmiento, A., Jiménez, S., &
Alemany, P. (2017). Automatic recognition of severity
level for diagnosis of diabetic retinopathy using deep
visual features. Medical & Biological Engineering
& Computing, 1-16. DOI 10.1007/s11517-017-
1638-6.
12. Doshi, D., Shenoy, A., Sidhpura, D., & Gharpure, P.
(2016, December). Diabetic retinopathy detectionusing
deep convolutional neural networks. In Computing,
Analytics and Security Trends (CAST), International
Conference on (pp. 261-266). IEEE. DOI:
10.1109/CAST.2016.7914977.
13. Quellec, G., Charrière, K., Boudi, Y., Cochener, B., &
Lamard, M. (2017). Deep image mining for diabetic
retinopathy screening. Medical Image Analysis, 39,178-
193.DOI: 10.1016/j.media.2017.04.012
14. Pratt, H., Coenen, F., Broadbent, D. M., Harding, S. P.,
& Zheng, Y. (2016). Convolutional neural networks
for diabetic retinopathy. Procedia ComputerScience,90,
200-205. DOI: 10.1016/j.procs.2016.07.014
15. Gargeya, R., & Leng, T. (2017). Automated
Identification of Diabetic Retinopathy Using Deep
Learning. Ophthalmology. DOI:
10.1016/j.ophtha.2017.02.008 ISSN
16. Colas, E., Besse, A., Orgogozo, A., Schmauch, B., Meric, N.,
& Besse, E. (2016). Deep learning approach for
diabetic retinopathy screening. Acta Ophthalmologica,
94(S256). DOI: 10.1111/j.1755-3768.2016.0635 24
17. Roy, P., Tennakoon, R., Cao, K., Sedai, S., Mahapatra, D.,
Maetschke, S., & Garnavi, R. (2017, April). A novel
hybrid approach for severity assessment of Diabetic
Retinopathy in colour fundusimages.In,2017IEEE14th
International Symposium on Biomedical Imaging (ISBI
2017) (pp. 1078- 1082). IEEE. DOI:
10.1109/ISBI.2017.7950703
18. Paul, S., & Singh, L. (2016, December).
Heterogeneous modular deep neural network for
diabetic retinopathy detection. In Humanitarian
Technology Conference (R10-HTC), 2016 IEEE Region
10 (pp. 1-6). IEEE. DOI: 10.1109/R10-
HTC.2016.7906821
19. Takahashi, H., Tampo, H., Arai, Y., Inoue, Y., &
Kawashima, H. (2017). Applying artificial intelligenceto
disease staging: Deep learning for improved staging of
diabetic retinopathy. PloS one, 12(6), e0179790. DOI:
10.1371/journal.pone.0179790
20. ElTanboly, A., Ismail, M., Shalaby, A., Switala, A., El‐Baz,
A., Schaal, S., ... & El‐Azab, M. (2017). A
computer‐aided diagnostic system fordetectingdiabetic
retinopathy in optical coherence tomography images.
Medical physics, 44(3), 914-923. DOI:
10.1002/mp.12071
21. Gulshan, V., Peng, L., Coram, M., Stumpe, M. C., Wu, D.,
Narayanaswamy, A., ... & Kim, R. (2016).
Development and validation of a deep learning
algorithm for detectionof diabetic retinopathyinretinal
fundus photographs. Jama, 316(22), 2402-2410. DOI:
10.1001/jama.2016.17216
22. Prentašić, P., & Lončarić, S. (2016). Detection of
exudates in fundus photographs using deep neural
networks and anatomical landmark detection fusion.
Computer methods and programs in biomedicine, 137,
281-292. DOI: 10.1016/j.cmpb.2016.09.018
23. Prentašić, P., & Lončarić, S. (2015, September).
Detection of exudates in fundus photographs using
convolutional neural networks. In Image and Signal
Processing and Analysis (ISPA), 2015 9th International
Symposium on (pp. 188-192). IEEE. DOI:
10.1109/ISPA.2015.7306056
24. Shan, J., & Li, L. (2016, June). A deep learning
method for microaneurysm detection in fundusimages.
In Connected Health: Applications, Systems and
Engineering Technologies (CHASE), 2016 IEEE First
International Conference on (pp. 357-358). IEEE. DOI
10.1109/CHASE.2016.12
25. Costa, P., & Campilho, A. (2017). Convolutional bag
of words for diabetic retinopathy detection from eye
fundus images. IPSJ Transactions on Computer Vision
and Applications, 9(1), 10. DOI: 10.1186/s41074-017-
0023-6
26. Vo , H. H., & Verma, A. (2016, December).NewDeep
Neural Nets for Fine-Grained Diabetic Retinopathy
Recognition on Hybrid Color Space.InMultimedia(ISM),
2016 IEEE International Symposium on (pp. 209-215).
IEEE. DOI 10.1109/ISM.2016.99