Review Article A Review on Recent Developments for ...downloads.hindawi.com/journals/scientifica/2016/6838976.pdf · Images of retina are taken by a device called fundus camera. Retinal

Review ArticleA Review on Recent Developments for Detectionof Diabetic Retinopathy

Javeria Amin Muhammad Sharif and Mussarat Yasmin

COMSATS Institute of Information Technology Department of Computer Science Wah 47040 Pakistan

Correspondence should be addressed to Mussarat Yasmin mussaratabdullahgmailcom

Received 14 December 2015 Revised 22 April 2016 Accepted 10 May 2016

Academic Editor Gary Lopaschuk

Copyright copy 2016 Javeria Amin et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Diabetic retinopathy is caused by the retinal micro vasculature which may be formed as a result of diabetes mellitus Blindness mayappear as a result of unchecked and severe cases of diabetic retinopathyManual inspection of fundus images to checkmorphologicalchanges in microaneurysms exudates blood vessels hemorrhages and macula is a very time-consuming and tedious work It canbe made easily with the help of computer-aided system and intervariability for the observer In this paper several techniques fordetectingmicroaneurysms hemorrhages and exudates are discussed for ultimate detection of nonproliferative diabetic retinopathyBlood vessels detection techniques are also discussed for the diagnosis of proliferative diabetic retinopathy Furthermore the paperelaborates a discussion on the experiments accessed by authors for the detection of diabetic retinopathy This work will be helpfulfor the researchers and technical persons who want to utilize the ongoing research in this area

1 Introduction

Diabetes is a very common disease worldwide It serves asa most common cause of blindness for people having ageless than 50 years It is a systemic disease which is affectingup to 80 percent of people for more than 10 years Manyresearchers acknowledged that 90 percent of diabetic patientscould be saved from this disease through an early diagnoseA person having diabetes is more prone to the risk of diabeticretinopathy (DR) [1] The blood supply towards all layersof retina is done through micro blood vessels which aresusceptible to unrestrained blood sugar level When a largeamount of glucose or fructose gathers in blood the vesselsstart crumbling because of insufficient distribution of oxygento cells Any blockage in these vessels leads to a severe eyeinjury As a result metabolic rate slows down and leadsto structural abnormality in vessels which intern DR [2]Microaneurysms are an earlier sign ofDRThis disease bringschanges in the size of blood vessels (swelling)The indicationsof DR include microaneurysms (MAs) exudates (EXs) andhemorrhages (HMs) as well as the abnormal growth of bloodvessels DR normally has two different stages named asproliferative DR (PDR) and nonproliferative DR (NPDR) [3]Occurrence of NPDR is when blood vessels in retina are

damaged and start leaking fluid onto it As a result retinabecomes wet and swollen Different signs of retinopathy existat this stage for example HMs MAs EXs and also interreti-nal micro vascular abnormalities (IRMA) PDR arises whennew abnormal blood vessels appear in various areas of retinaIt is a complex case of DR that may cause impaired vision[4] DR is a progressive disease and its detection at an earlystage is very crucial for saving a patientrsquos vision this requiresregular screening An automated screening system forDR canhelp in reducing the chances of complete blindness due to DRalong with lowering the workload on ophthalmologists ForDR screening a computer-aided diagnostic (CAD) systemis developed for differentiating a retina with possible DRfrom a normal retina [5ndash7] Figure 1 shows the symptomsfor different stages of DR The paper is organized as followsSection 2 describes different publicly available databases thatare used for the detection of DR Section 3 mentions variousperformance measures that are taken into account for systemevaluation Section 4 gives an overview of different methodsbeing used for the detection of DR Section 5 highlightsdifferent techniques that are applied for the screening ofDR Section 6 details retinal imaging techniques Section 7focuses on methods for DR detection using CAD for thedetection of DR different stages

Hindawi Publishing CorporationScientificaVolume 2016 Article ID 6838976 20 pageshttpdxdoiorg10115520166838976

2 Scientifica

MAs HMsEXs

(a)

Abnormal growth of vessels

(b)

Figure 1 Stages of diabetic retinopathy (a) Signs of NPDR (b) Signs of PDR

2 Publicly Available Databases

Images of retina are taken by a device called fundus cameraRetinal fundus images (RFI) is the name given to theseimages This camera takes images of the internal surface ofretina posterior pole macula optic disc and blood vesselsImage acquisition is a leading step for medical diagnosisFor research on medical fundus images widely accessibleresources are available as shown in Table 1 Some benchmarkdatabases are openly available for the assessment of algo-rithms introduced for the computerized screening and anal-ysis of DR The purpose of databases is to check the strengthof automatic screening of DR and then compare the resultswith current techniques Seven datasets are available openlyincluding DRIVE STARE DIARETDB e-ophtha HEI-MED Retinopathy Online Challenge (ROC) and Messidor

3 Performance Measures

For the diagnosis of DR in its early stage the retinal imagecaptured using some fundus camera needs to undergo prepr-ocessing before applying further algorithms of image proces-sing Many preprocessing techniques such as contrast adjust-ment average filtering adaptive histogram equalizationhomomorphic and median filtering are applied on retinalimages in the gold standard database After an algorithm isapplied to a retinal image mean square error (MSE) andpeak signal to noise ratio (PSNR) are further calculated toanalyze the performance of an algorithm PSNR is mostlydenoted regarding logarithmic decibel value A higher valueof PSNR means that the processed image is of higher andbetter quality than the actual image In medical treatmentthe medical contribution data is frequently divided into twotypes data where the disease is present and data where thedisease is not diagnosedThe level of correctness of treatmentis reviewed by the sensitivity and specificity measures Inmedical research fundus images which are common in DRare calculated via sensitivity and specificity of each imageThe higher sensitivity and specificity values enhance thetreatment True positive (TP) denotes the total number oflesion pixels and true negative (TN) denotes the nonlesionpixels False positive (FP) denotes the number of nonlesionpixels that are detected wrongly by the algorithm Likewisefalse negative (FN) indicates the number of lesion pixels

that are not identified by the algorithm Table 2 showsperformance metrics

4 Methods for Detection of DiabeticRetinopathy

DR is a main reason of blindness DR development is at dif-ferent rates in different persons because of the two importantvision pressuring difficulties diabetic macular edema (DME)and proliferative retinopathy That is why there is a hugedemand for the latest technologies andmethodologies to ana-lyze DR efficiently and correctly in its early stage Nowadaysthe research community has givenmany techniques shown inFigure 2 for early detection of DR that is conferred here

5 Screening of Diabetic Retinopathy

Eye illnesses for example DR and DME are the most widelyrecognized reasons for irreversible vision loss in people withdiabetes Just in the United States alone health care andrelated costs identified with eye maladies are assessed atnearly $500M Besides the prevalent cases of DR are reliedupon to become exponentially influencing more than 300Mindividuals around the world by 2025 Early discovery andtreatment of DR and DME play a major role in prevent-ing adverse effects such as blindness Optical CoherenceTomography (OCT) imaging is ideal for uncovering DME onaccount of its extended axial resolution and extended retinalscan coverage spatial domain optical coherence tomography(SD-OCT) revealed DME-related changes of the photore-ceptors external restricting film Bruchrsquos layer and retinalpigment epithelium layer Moreover SD-OCT perceivedrelated epiretinal layers changes in the retinal decay andvitreoretinal interface The issues of automatic classificationof SD-OCT information for automatically recognizable proofof patients with DME versus ordinary subjects are tendedas well The proposed technique depends on Local BinaryPattern (LBP) features to portray the texture of OCT picturesand contrast diverse LBP features extraction approaches withthe process of single signature for the entire OCT volumeTest results with two datasets of separately 32 and 30 OCTvolumes demonstrate that regardless to utilizing low or highstate representations features obtained from LBP texture

Scientifica 3

Table1Databases

forfun

dusimages

Nam

eIm

agea

cquisition

Num

bero

fimages

Resolutio

nUses

DRIVE[8]

3-CC

Dcameraw

ith45-fo

ldview

20colorfun

dustestin

gim

ages

20colorfun

dustrainingim

ages

768times584

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

Image-Re

t(DIARE

TB0

DIARE

TB1)[9]

50-fo

ldview

DIARE

TB0total130

images

inwhich

20im

ages

aren

ormaland110

with

DR

DIARE

TB1total89im

agesfive

images

aren

ormaland84

images

with

DR

1500times1152

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

Messid

or[10]

3CCD

cameraa

t45-fold

view

1200

images

1440times9602240times1488and

2304times1536

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

Retin

opathy

OnlineC

hallenge

[11]

Cano

nCR

5-45-N

Mcameraa

t45-fo

ldview

100digitalfun

dusimages


1389times1383

Microaneurysm

sdetectio

n

e-op

htha

EX[12]

OPH

DIATcopy

Tele-m

edicalnetwork

Itcontains

47im

ages

with

exud

ates

and

35im

ages

with

nolesio

n2048times1360

Exud

ates

detection

e-op

htha

MA[12]

OPH

DIATcopy

Tele-m

edicalnetwork

Itcontains

148im

ages

with

microaneurysm

sorsmallh

emorrhages

and233im

ages

with

nolesio

n2048times1360

Microaneurysm

sdetectio

n

STARE

[13]

TopconTR

V50

fund

uscameraa

t35-fo

ldview

400im

ages

605times700

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

HEI-M

ED[7]

Zeiss

Visucam

PROfund

uscamera

at45-fo

ldview

169im

ages

inwhich

115im

ages

are

abno

rmaland54

images

areh

ealth

y2196times1958

Exud

ates

detection

4 Scientifica

Table 2 Performance metrics

Measures Description

Peak signal to noise ratio (PSNR) 20 log10(MAX

119868) minus 10 log

10(MSE) Measures quality of image

Sensitivity or true positive rate(TPR)

TPR = TPP=

TPTP + FN Measures the ratio between TP and FN

False positive rate (FPR) FPR = FPN=

FPFP + TN

= (1 minus SPC) Measures the ratio between FP and TN

False negative rate (FNR) FNR = FNTP + FN

= (1 minus TPR) Measures the ratio between FN and TP

Specificity (SPC) or true negativerate (TNR)

SPC = TNN=

TNTN + FP Measures the ratio between TN and FP

Accuracy (ACC) ACC = (TP + TN)TP + FP + FN + TN

The degree to which the result of ameasurement calculation or specificationconfirms the correct value or a standard

Area under curve (AUC) 119860 = int

minusinfin

infin

TPR (119879) FPR1015840 (119879) d119879 How much system is sensitive to detect thedesired output

Diabetic retinopathy

Screening of diabetic retinopathy

Retinal imaging

Diabetic retinopathy byCAD

Nonproliferative

Proliferative

Microaneurysm

Macular edema

Exudates

Hemorrhages

Blood vessels abnormalities

Figure 2 Methods for detection of diabetic retinopathy

have profoundly discriminative power [14 15] A novelstrategy is recommended that uses noncalibrated numerousperspective fundus pictures to dissect the swelling of maculaThis development empowers the discovery and quantitativeestimation of swollen zones by remote ophthalmologistsThisability is not accessible with a single image and inclinedto error with a stereo fundus camera Likewise exhibitautomatic algorithm to quantify features from the recreatedimage which are helpful in POC robotized conclusion ofearly macular edema for example before the presence ofexudation The proposed method is divided into three stepsinitial a preprocessing system at the same time enhancesthe dim microstructures and the procedure of macula andbalances the image second all accessible perspectives areenlisted utilizing nonmorphological sparse features at longlast a thick pyramidal optical flow is computed for each oneof the images and statistically consolidated to build a naıveheight map of macula Results are demonstrated on threearrangements of synthetic images and two arrangements ofreal world images These preparatory tests demonstrate thecapacity to deduce a minimum swelling of 300 120583m and tocorrelate the recreation with the swollen area [16]

6 Retinal Imaging

Optical systems are enhanced by utilizing the method ofadaptive optics (AO) It is accomplished by reducing thedistortion in front wave effect Diseased eye and normaleye can be evaluated for spacing mosaic and photographycell thickness through AO retinal imaging With the helpof retinal imaging the inflammatory diseases are monitoredfor the posterior segment of fundus photography Fundusfluorescein angiography (FA) and OCT are the most com-monly used techniques for retinal imaging These imagingmodalities are helpful in the therapy of patients with inflam-matory conditions for the correct diagnosis of posterior pole[17] FA is an invasive technique which utilizes fluoresceincolor and OCT is a costly imaging strategy contrasted withfundus photos Besides utilizing fundus pictures DME canbe automatically diagnosed DME grading method is utilizedfor telescreening Henceforth DME grading method appliedon fundus image modalities is reliable and inexpensivetechnique contrasted with biomicroscopy OCT and FAmodalities [18] En face regular or improved multicolorscanning laser ophthalmoscopy (SLO) images gained with

Scientifica 5

HRA2 permit a superior perception of epiretinal films forpreoperative assessment contrasted with SD-OCT based enface thickness guide or pseudoshading images obtained withOptomap while infrared or FA images are slightly suitableto portray epiretinal layers [19] Three-dimensional regionalstatistics are used to detect disease macular area using OCTimages The proposed method is tested on five patientswith retinal malady in OCT images which indicates 807accuracy for the anomalous range [20]

7 Diabetic Retinopathy Detection byComputer-Aided Diagnostic System (CAD)

Theobjective of computer-aided diagnosis is to recognize DRand normal images in utilizing features like area of EXsMAsveins texture HMs node points and so forth [34 35]

In order of two classes (DR andnormal) forDR screeningdevice was produced by (Usher et al [23] Sinthanayothin etal [36] Aptel et al [21] Reza and Eswaran et al [25] Gardneret al [24] Kahai et al [22] Osareh et al [37] and Quellec etal [38]) utilizing clinical components in particular EXs veinsMA CottonWool Spots (CWS) and HMs Dark lesions weresegmented out using moat operator and EXs were extricatedusing recursive region growing (RRGT) and adaptive inten-sity thresholding (AIT) [23] Quellec et al [38] have appliedoptimal filters to extricate MAsThe artificial neural network(ANN) [23 24 36] and Bayesian outline work [21] wereused for classification Their methods obtained specificity of463 and sensitivity of 951 [23] Support vector machine(SVM) kernel classifiers are employed in CAD frameworkto discover the absence or presence of DR The anticipatedCAD framework has managed the classification issues in DR[26] Computer-aided diagnosis has assumed a fundamentalpart in medical industries [39] To recognize DME DRand essential injury an algorithm in view of examination ofmechanized fundus photo was proposedWhile screening forDR this algorithm is an enhanced substitution for the fundusphoto manual examination which expends a considerablemeasure of time If this framework is operated by the doctorsit will allow significant time saving hence enhancing thetime spent on a mass screening program [27] In medicinalimaging contextual information performs an importantrole Bright lesions detection and discrimination has beenperformed through contextual information in fundus imagemodalities Diagnosis of coronary calcifications and hard EXsin CT scan is performed through this contextual informationIn the spatial relation high-level contextual features are usedto explain the context Contextual CAD framework is anoutflanked loomwhen contrastedwith local CAD framework[40] Several authors have suggested CAD methods forthe detection of different stages (NPDR and PDR) of DRthat are encountered and described in this section Table 3summarizes these CAD methods

71 Segmentation and Localization of Optic Disc The opticdisc (OD) is a round region in the back of eye where retinalnerve fibers gather to frame the optic nerve OD is sometimescalled optic nerve head (ONH) since it is the leader of opticnerve as it enters eye from the brain It is found marginally

Optic disc

Figure 3 Optic disc in fundus image

to the nasal side of globe OD is known as the blind sidesince it contains no photoreceptors In this way any lightcentered on OD can neither be changed over into sensoryimpulses nor sent to the brain for elucidation Figure 3 showsOD in a fundus image For the detection of DR first stepis segmentation and localization of OD because its colorintensity and contrast are the same as other features of retinalimages Many authors have suggested CAD methods for thesegmentation and localization of OD that are described inTable 4

For automatically finding ONH the strategy of Gaborfilters in fundus image is used as the ONH center whichis close to the focal point of retinal vessels and the phaseportrait analysis [60 61] New method is proposed for thelocalization and detection of OD The proposed techniquegives better results [62] Histogrammatching method is usedfor OD localization [63] Morphological filtering strategiesand watershed transformation are used for OD detection andlocalization The proposed method has been tested on 30color fundus images Subsequently achieved mean predictivevalues and sensitivity were 924 and 928 respectively[64] The proposed method is used for localizing the ODcenter that depends on corners and bifurcations attainedwith Harris corner detector The achievement rate is 8765for STARE 975 for DRIVE and 978 for local datasetrespectively [65] Macula centers and OD are detected usingradial symmetry method [66 67] DR is diagnosed by usingthe method of retinal extraction In automatic screening ODis detected at very low cost Locating the center of OD isdifficult because the color brightness and contrast are similarto CWS and EXs [68] Boundary tracing technique is usedfor locating the OD boundary [69] Multilevel 2D waveletdecomposition method is used for the localization of ODHRF database is utilized for the assessment of results withReceiver Operating Characteristic (ROC) curve and 95accuracy is achieved for the localization of OD [29] For ODdetection and location a new method is proposed in thelight of histogram approaches and clustering The methodis evaluated with Messidor dataset which demonstrates thatit can localize the OD accurately even in blurred images[30] A line operator is proposed to capture round brilliancestructurewhich assesses the image shine variety along variousline segments of particular orientations that go through everyimage pixel of retina The orientation of line segment withbasegreatest variety has a particular example that can beutilized to find the OD precisely Suggested line operatoris tolerant to a normal OD identification various sorts oflesions in retinal imaging modalities with an achievement

6 Scientifica

Table3CA

Dmetho

dsford

iagn

osisof

DR

Algorith

mIm

agep

rocessing

techniqu

esDatabase

Color

space

Sensitivity

Specificity

AccuracyAUC

Lesio

nsdetection

Apteletal[21]

Sing

le-field

nonm

ydria

tic

single-field

mydria

tic

three-field

nonm

ydria

tic

three-field

mydria

tic

79patie

nts(158eyes)

Grayscale

7790

92

97

9998

97

98

082090090095

Both

NPD

RandPD

R

Kahaietal[22]

Decision

supp

ort

syste

m(D

SS)

143im

ages

Louisia

naState

University

Eye

Center

Grayscale

100

67

mdashNPD

R

Usher

etal[23]

Neuraln

etwork

1273

consecutive

patie

ntsSt

Thom

asHospital

Grayscale

948

mdashmdash

NPD

R

Gardn

eretal[24]

Back

prop

agation

neuralnetwork

200diabetic

and101n

ormal

images

ofprivate

hospita

l

Grayscale

884

835

mdashBo

thNPD

RandPD

R

Reza

andEswaran

[25]

Rulebasedcla

ssifier

STARE

Green

channel

972

100

97

NPD

R

Ann

ieGrace

Vimalaa

ndKa

jamoh

ideen[26]

Cost-e

ffective

compu

ter-aideddiagno

stic

syste

m

Privatee

yeHospital

HSV

916

905

912

NPD

R

Dup

asetal[27]

Automated

fund

usph

otograph

analysis

algorithm

sMessid

orGrayscale

839

727

mdashNPD

R

Ashrafetal[77]

LocalB

inaryPatte

rn

SVM

DIARE

TB1

Green

channel

8748

8599

8615

087

NPD

R

Scientifica 7

Table4Differentm

etho

dsford

etectio

nof

optic

disc

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Accuracy

Rathod

etal[29]

Multilevel2D

wavele

tHistogram

Equalization

HRF

Green

channel

95

Sekara

ndNagarajan

[30]

ODlocalizationbasedon

cluste

ringandhisto

gram

approaches

Messid

orRG

B9958

LuandLim

[31]

Circular

transfo

rmation

lineo

perator

DIARE

TDB0

DIARE

TDB1

DRIVE

STARE

CIEL

AB

974

Acharyae

tal[32]

Otsu

Gradientvectorfl

ow(G

VF)

snakeAtanassov

Intuition

isticFu

zzyHiston

(AIFSH

)Ka

sturbaM

edicalCollege

Grayscale

100

Trucco

etal[33]

Binary

vesselmasks

ared

evelo

pedwith

inVA

MPIRE

softw

are

HRIS(H

igh-Re

solutio

nIm

age

Set)

VDIS

(VascularD

iseaseImage

Set)

Grayscale

957

921

8 Scientifica

Figure 4 Fundus image with exudates

of 974 accuracy [31] A novel algorithm is proposed forthe segmentation of OD in images of retina while makinguse of Atanassov Intuitionistic Fluffy Histon (AIFSH) basedsegmentationmethod Columnwise neighborhood operationand pixel intensity of OD are utilized to find and separatethe OD [32] A novel method is proposed depending on asymmetry transform and in painting which is applied verycompetitively in tests with a publicly available and localdataset [33] A summarization of different OD detectionmethods is shown in Table 4

72 Segmentation of Soft and Hard Exudates (NPDR) EXsare one of the primary signs of DR which can be preventedwith an early screening process Some existing work for thedetection of EXs is described below

A novel approach is proposed to detect DR automaticallyfrom digital fundus images Digital images play a significantrole in the identification of DR by calculating various colorspaces in segmented region DR is detected by fuzzy set usingfuzzy logic reasoning [86] In developed countries visualimpairment is a significant reason of DR The disease can bedetected to have the assistance of fundus pictures identifiedwith DR lesions For this reason a method is proposed usinga neural network The proposed methodology is completelyautomatic after the configuration of classifiers [87] EXs canbe detected using multispace clustering method [41 50]Figure 4 presents a fundus image with the symptoms of EXs

The proposed method is reliable for the identification ofEXs in retinal images Morphological operators and adaptivethresholding method are utilized in the computation ofnoise map distribution Contrast changes and nonuniformillumination method are used for the detection of correctEXs [42] The proposed method is used for accurate seg-mentation of EXs [88] For the diagnosis of bright lesions119870-means clustering method is applied [89] A novel methodis proposed utilized neural network to minimize DR withhigh accuracy [43] Both KNNFP and WKNNFP classifiersare used to detect the EXs but WKNNFP shows betterresults as compared to KNNFP [44] Bright lesion can bedetected using Gabor filter [90] Mathematical morphologyalgorithms and 119870-means are utilized for the identificationof bright lesions [45] EXs can be recognized with the helpof gray level variation and their contours are determinedusing the morphological reconstruction methods [91] Haarwavelets transform is used for the hard EXs segmentation

followed by 119870-nearest neighbor classification method Theproposed method is tested on four databases of fundusimages among them obtained sensitivity is 3714 21871250 and 2547 for MISP (Medical Image and SignalProcessing Research Center) DIRETDB0 DIREDB1 andSTARE database respectively Likewise the specificity is 0for MISP and 1 for remaining databases [46] The sug-gested technique utilizes several image processing methodsincluding image thresholding and median filtering withan aim to discover hard EXs The suggested techniqueshowed specificity of 9685 and sensitivity of 9725 [2]A novel technique is proposed for the discovery of brightlesions in color retinal images Intellectual decision supportsystem is utilized for DR detection The texture and colorfeatures are applied for distinguishing between non-EXsand EXs pixels Initially edge discovery and morphologicalprocess are used for the segmentation of OD Secondly forattaining texture features from the area of retina color andlaws texture energy measures are performed Afterwardsan intelligent classifier Fuzzy SVM has been utilized todiscover pathological regions in color fundus images [47]The EXs discovery technique comprises two stages fineand rough EXs segmentation Rough segmentation has beenapplied using columnwise neighborhoods and morphologyoperation whereas morphological reconstruction method ispractical for fine segmentationThe suggested technique usedretinal image database fromMalaysia Sungai BulohHospitalOrganized with other appropriate retinal features extractionand classification technique this segmentation technique canform the basis of an easy and fast diagnostic support tool forDR which will provide a great benefit regarding better accessto mass screening people for risk or existence of diabetes[48] The new technique is proposed for automatic discoveryof human fundus image by the submission of digital imageprocessing Circular bit plane slicing and Hough transformare applied for OD localization in the proposed techniquewhereas for the extraction of EXs morphological operationsare used The suggested technique is a novel method It hasa sensitivity of 9362 and an accuracy of 88 [49] Bagof Words algorithm is used to make a system which playsthe role of both the case based reasoning (CBR) systemand decision support system (DSS) to solve the problemof bright lesion segmentation [51] For the segmentationof EXs dynamic region growing method is proposed Themethod is tested on several images of retina and the resultsdemonstrate that the method does better than the earliersuggested techniques [52] Novel measurable atlas basedtechnique is proposed for the segmentation of EXs Any testfundus picture is initially distorted on Atlas coordinate andafter that a distance map is obtained with the mean atlaspicture An analysis of openly accessible HEI-MED datasetshows great presentation of the technique On the FROCcurve 35 nonlesion localization fraction and 825 local-ization fraction are obtained The technique is additionallycontrasted with couple of latest reference strategies [92]Hard EXs can be characterized by the DME Novel featuresset based on color and wavelet decomposition are used forDME detection Classifier is trained using these featuresto automatically analyze DME through the occurrence of

Scientifica 9

Figure 5 Macula exudates and optic disc in fundus image

exudation Another freely accessible HEI-MED database isthere with ground-truth information containing 169 patientsOur calculation acquired an AUC somewhere around 088and 094 relying upon the datasetfeatures utilized [93]

A brief summary of methods for the detection of EXs isgiven in Table 5

73 Macular Edema Detection (NPDR) Macula is the centerpoint of vision where light is focused In a fundus imagemacula appears at the center of retina with a blackish colordue to an excess of melanin in its composition It interpretsthe images and sends them to brain Its size is 3mm Iflipids and proteins start accumulating near or on the maculait can become as dark lesions these are called macularedema Figure 5 shows EXs macula and OD in a fundusimage Many researchers have proposed CAD systems for thedetection of DME that are described in Figure 5

DME is a fundamental reason of visual harm Thepathogenesis of macular edema seems to be multifactorialLaser photocoagulation is the standard of care for macularedema [94 95] Surgical therapy and pharmacologic methodare used to handle DME in diabetic patients [96] Thequalitative OCT and clinical examinations are required tobe performed on monthly basis to check the anti-VEGF(vascular endothelial growth factor) for the maximizationratio of visual acuity gain along with the requirement of someinjections [97 98] The treatment by injections is safe overtwo years and effective for DME DME treatments includeVEGF drugs There is no damage to retina by the MicropulseDiode Laser (MPL) treatment Retinal pigment epithelium(RPE) is used for different MPL stimulation ETDRS photo-coagulation group is compared with laser group and foundthat retinal sensitivity has been increased [99 100] Gaussianmixture model (GMM) based classifier and detailed featureset are utilized for perfect and accurate detection of maculathis nominatedsubmitted system consists of a novel methodBoth SVM and ensemble of GMM classifiers are used for anaccurate detection of EXs which ultimately leads to perfectclassification of retinal image in different stages of macularedema The same system has got mean value of 959973 and 968 for specificity sensitivity and accuracyrespectively [101] Raja and Ravichandran [102] explain a wayto autolocalize the fovea center in retinal fundus imagesThis

Microaneurysms

Microaneurysms

Figure 6 Microaneurysms in retinal image

method is specifically based on mathematical morphologybeside the information of other anatomic structures such asblood vessels and OD Firstly the vascular structure and ODcenter are extracted and then morphological operations areemployed on the gray scale image of green channel for foveacandidatesrsquo selection The candidates satisfying area densityand distance criteria are considered for the final stage Andthere the candidate having lesser vessel pixels is selectedas fovea region The method was evaluated on two publiclyavailable STARE and DRIVE databases It was able to obtain100 of fovea localization accuracy on DRIVE database with288 seconds average computation time

74 Microaneurysms Detection (NPDR) An early symptomof DR is MAs MAs diagnosis is significant in early detectionofDRMAs are themajor symptomofNPDRand are initiatedby the principal dilatations of thin blood vessels Figure 6shows MAs in retinal image Several proposed CAD systemsfor the detection of MAs are discussed in Figure 6

MAs are of small dimensions nearly red in color andround [103] Dynamic thresholding and multiscale correla-tion filtering (MSCF) method are used for MAs detectionThe proposed method contains two levels coarse level (MAscandidate detection) and fine level (true MAs classification)The method was tested on two publicly available datasetsnamely ROC and DIARETDB1 databases [104] 119870-nearestneighbor classifier (119870NN) is used for the detection of MAs[105]Morphological operators are applied forMAs detectionin fundus images The method obtained 9998 accuracy8161 sensitivity 6376 precision and 9999 specificity[53] An ensemble-based framework is nominated to improveMAs detection [54] The proposed method comprises twostages In the first stage preprocessing is performed usingfractal analysis of retinal vascular structure The principlestage contains picture preprocessing and fractal analysis ofretinal vascular structure If fractal examination distinguishesan abnormal image from a normal one this enhances theeffectiveness of a computerized screening program Iden-tification of MAs distinctive shape as an abnormal retinalpicture through morphological reconstruction methods andcanny edge detection is an objective of second stage Theproposed calculation has been performed on an arrangementof 89 fundus pictures from the accessible database The

10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE

119871lowast119906lowastVlowast

HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages

Figure 7 Hemorrhages in fundus image

applied method achieves operating sensitivity and highestspecificity as 895 and 821 respectively [55] Contrastenhancement approach with Singular Value Decomposition(SVD) method is applied to fundus images After it Hessian-based candidate selection method is used for the detectionof MAs For every candidate region intensity normalizedradon transform and robust low-level blob descriptors areknown as SURF and take out to characterize MAs candidateregions Then classification is performed along with SVMclassifier which has been trainedwith tenmanually annotatedtraining images Presentation of complete system is estimatedon ROCdatabase [56] Contrast LimitedAdaptiveHistogramEqualization (CLAHE) method is used to enhance contrastof images for the detection of lesions [57] The localizationand detection of MAs are based on three steps candidatesrsquoselection segmentation of MAs and a final performanceevaluation New radon cliff operator is proposed which is anactual contribution to the field [58] The dynamic multipa-rameter template (DMPT) matching scheme is applied forthe detection of MAs that is more accurate as compared toconventional schemes [59]

A summary of MAs detection methods is presented inTable 6

75 Hemorrhages (NPDR) HMs occurs when blood outflowsfrom retinal vessels Figure 7 shows HMs in a fundus imageMany authors suggested CAD systems for detection of HMswhich are listed and discussed in Figure 7

The proposed technique used HSV color space by non-linear curve for changing the brightness of fundus imagesBrown regions are highlighted using gamma correction ineach blue green and red bit image Then histogram of eachblue green and red bit image was extended Brown regionsrepresented HMs and blood vesselsThen density estimationis applied to find brown regions False positives were elim-inated via 45-feature examination The proposed method istested on 125 fundus images along with 90 normal imagesand 35 images with HMs [70] In fundus image a novel splatfeature division method is proposed for retinal HMs detec-tion A new method is proposed for the preprocessing andfalse positive elimination A classifier is prepared with splat-based skilled observations and expanded on openly accessibleMessidor dataset [71] Automated Decision Support System(DSS) is developed for the detection of HMs and MAs infundus images The severity level of DR is determined by

Figure 8 Abnormal blood vessels

the location and number of HMs and MAs The method istested on 98 fundus images Correspondingly experimentaloutcomes show that 8753 and 8431 sensitivity and 9508and 9363 specificity values were obtained by the proposedsystem for HMs and MAs discovery [72 73] Sudha andThirupurasundari [74] suggested an automated system todiscoverDR from retinal images In thismethod after prepro-cessing texture features are taken out from retinal images todiscover abnormal images Afterwards the abnormal imagesare treated to localize and classify the problem of HMs andEXs Dynamic thresholding method is used for the detectionof HMs in retinal images Experimental results show thatthe HMs are discovered with good accurateness in retinalimages [75] A novel hybrid classifier is proposed for thediscovery of retinal lesionsThe suggested method comprisespreprocessing taking out of feature set formulation andclassification of candidate lesions The system is measuredusing standard fundus image databases with the supportof performance parameters known as specificity accuracysensitivity and ROC curves for statistical implementation[76] For HMs detection concentration is on the examinationof texture micro-patterns of regions of interest (ROIs) whichare concerned areas in an image of retina Texture micro-structures of ROIs are examined via LBP for their expla-nation Lastly SVM is applied to indicate whether an ROIconsists of HMs or not [77] A brief overview of commonlyused methods for HMs detection is given in Table 7

76 Blood Vessel Detection (PDR) PDR occurs when there isformation of some new abnormal blood vessels in differentregions of retina It is an advanced stage ofDR andmay lead tocomplete blindness Hence for PDR diagnosis blood vesselsdetection is an important task Figure 8 shows abnormalblood vessels Some CADmethods for the detection of bloodvessels are described in Figure 8

DR is described against the classification and detectionof changes in time series as presented in fully automatedapproach [28 106] This method consists of the followingsteps (1) illumination from instrument and patient visit(2) dust particle imaging (3) training data and (4) align-ment and segmentation error of retinal vasculature foveaand OD Pathologies are extracted automatically and theirrobustness is achieved by an algorithm The technique ispresented for the bifurcation and geometric model withoutthe intervention of user [107] For blood vessels segmen-tation the proposed method applied subpixel root MSEwith the adoption of preprocessing MultilevelThresholding

12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]

Medianfiltercontrastenhancem

entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]

Ensemble-basedmicroaneurysm

sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica

H-maxima transformation and Binarization From this tech-nique accuracy rate of 965 is obtained [108] A novelmethod is proposed utilizing Gumbel probability distribu-tion function as its kernel and matched filter The methodgives accurate results for blood vessels segmentation [109]Similarly the methods of branches filtering and Max-Treerepresentation are utilized for the segmentation of vesselswith 9395 success rate [110] Fuzzy 119870-median and lengthfilter (FKMED) and matched filter methods are applied forblood vessels segmentation in fundus images The methodachieved sensitivity and specificity as 7931 and 9643 [111]Another derived methodology is Gaussian Intensity Distri-bution (GID) model utilized for blood vessels segmentation[112] Morphology method and canny algorithm are alsoused for blood vessels segmentation The proposed methodis tested on DRIVE dataset [113] The authors made useof matched filter and threshold probing for blood vesselssegmentation Grid division nonuniform enhancement andoptimal multithreshold method are used for blood vesselssegmentation [114] Analysis of DR on edge detectionmethodis based on content-based image retrieval (CBIR) frameworkPreprocessing methods are subjected to normal and abnor-mal images of retina to enhance the edge information Kirschtemplate and canny edge are two different methods that arepractical for blood vessels segmentation in the diagnosis ofretinal abnormalities The kirsch edges are useful in CBIRsystem [115 116] Fraz et al [78] introduced an ensemblesystem of boosted decision trees where bagged is used asfeature vectors based on orientation analysis of gradient vec-tor field Gabor filter responses line strength measures andmorphological transformation To handle the pathologicalretinal image the feature vector encodes information Forthe segmentation of blood vessels an automatic method issuggested The submitted method is based on the fact thatby varying the length of a basic line detector line detectorsat continuously changing scales are achieved In order toperform final segmentation for every image of retina lineresponses are linearly combined at varying scales so that thestrength and drawbacks elimination of each line detectorare maintained Both quantitative performance and quali-tative performance of this method were evaluated on threepublicly available DRIVE REVIEW and STARE datasets[79] For pixel classification based method to segment bloodvessels of retina linear discriminant analysis is describedin detail In this method vesselness measure of a pixel isdefined by Gabor filter responses and the feature vectoris comprised of a modified multiscale line operator [80]A concise methodology is introduced for segmenting theretinal vasculature in color fundus images [81] Lattice NeuralNetworks with Dendritic Processing (LNNDP) method isused to solve this problem [82] Pixel classification is doneby a neural network (NN) scheme [83] Multidirectionalmorphological top-hat transform with rotating structuringelements and enhanced multiscale line detector are utilizedfor blood vessels detection [84] Morphological operators areused for detecting blood vessel tree In retinal fundus imagesidentification of abnormal spots is done more accuratelyafter vessel detection Experimental results are taken fromNikookari database which is consisting of 40 fundus images

The method achieved 8582 average sensitivity and 9998average specificity [85]

A summary of blood vessels detection methods is pre-sented in Table 8

8 Future Directions

Inmedical image processing automatic diagnosis ofDR fromdigital fundus images has been a dynamic research for along time The research interest is justified by considerablereductions in health care costs and tremendous potential fornew products in medical industry There are certain areas inthis field that need improvement such as the determinationof OD boundary which is tough in two-dimensional retinalimages because of blur edges A difficult task to perform isblood vessels extraction Every subject has different ONHstructure Therefore there is not a single technique thatcan cover all these problems There is still need to proposemore efficient algorithms for the identification of DR detec-tion related structure and retinal changes Because of theexpanding predominance of diabetes mellitus demand fordiabetic retinopathy screening stages is steeply expandingEarly location and treatment of DR are essential public healthintercessions that can incredibly diminish the probabilityof vision loss Current DR screening programs commonlyutilize retinal fundus photographywhich depends on talentedreaders for manual DR evaluation However this is labor-intensive and suffers from inconsistency across sites Hencethere has been the latest proliferation of computerized retinalimage investigation programming that might mitigate thisweight cost viably Moreover current screening programsgiven 2-dimensional fundus photography do not properlyscreen for DME OCT is turning out to be progressivelyperceived as a reference standard forDME evaluation and cangive an economical solution for enhancing DME discoveryin vast scale DR screening programs Current screeningsystems are additionally not able to picture the peripheralretina and require pharmacological pupil dilation ultra-widefield imaging and confocal examining laser ophthalmoscopywhich addresses these disadvantages have excessive potential

9 Conclusion

Diabetic retinopathy cannot be cured To prevent vision losslaser analysis (photocoagulation) is usually very effective if itis done before it adversely harms the retina Provided that thestern destruction of retina has not been done vision can beimproved by the surgical elimination of vitreous gel (vitrec-tomy) In proliferative diabetic retinopathy at times an anti-inflammatory medicine or antivascular endothelial growthfactor medication injection can help in the new blood vesselscontraction process Since symptoms cannot build up untilthe disease turns into the stern initial discovery via regularscreening is essential Nonproliferative diabetic retinopathycontains early indications of DR and it is extremely critical torecognize and analyze DR at its initial stages If a person withdiabetes gets legitimate eye mind consistently and treatmentwhen fundamental DR will once in a while cause all outblindness In this study of DR a large portion of work is done

Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo

rmrotatingstructuringele

ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica

to discover hemorrhages microaneurysms and exudatesdiabetic macular edema and abnormal new blood vessels asthey are indications of the vicinity of retinopathy in fundusimages This study helps in the detection of retinopathy atan early stage timely treatment of this disease will preventpermanent vision lossThepaper discussed experiments doneby authors for the detection of diabetic retinopathyThisworkwill be useful for technical persons and researchers who needto use the ongoing research in this area

Competing Interests

The authors declare that they have no competing interests

References

[1] T Kauppi V Kalesnykiene J-K Kamarainen et al ldquoTheDIAR-ETDB1 diabetic retinopathy database and evaluation protocolrdquoin Proceedings of the 18th British Machine Vision Conference(BMVC rsquo07) pp 151ndash1510 September 2007

[2] D Kayal and S Banerjee ldquoA new dynamic thresholding basedtechnique for detection of hard exudates in digital retinalfundus imagerdquo in Proceedings of the 1st International Conferenceon Signal Processing and Integrated Networks (SPIN rsquo14) pp 141ndash144 February 2014

[3] C Agurto V Murray E Barriga et al ldquoMultiscale AM-FMmethods for diabetic retinopathy lesion detectionrdquo IEEE Trans-actions on Medical Imaging vol 29 no 2 pp 502ndash512 2010

[4] G Mahendran R Dhanasekaran and K N Narmadha DevildquoIdentification of exudates for Diabetic Retinopathy based onmorphological process and PNN classifierrdquo in Proceedings ofthe 3rd International Conference on Communication and SignalProcessing (ICCSP rsquo14) pp 1117ndash1121 IEEE MelmaruvathurIndia April 2014

[5] C Agurto V Murray H Yu et al ldquoA multiscale optimizationapproach to detect exudates in the maculardquo IEEE Journal ofBiomedical and Health Informatics vol 18 no 4 pp 1328ndash13362014

[6] A F Aqeel and S Ganesan ldquoAutomated algorithm for retinalimage exudates and drusens detection segmentation andmeasurementrdquo in Proceedings of the IEEE International Confer-ence on ElectroInformation Technology (EIT rsquo14) pp 206ndash215Milwaukee Wis USA June 2014

[7] P M Rokade and R R Manza ldquoAutomatic detection ofhard exudates in retinal images using haar wavelet transformrdquoInternational Journal of Application or Innovation in Engineeringamp Management vol 4 no 5 pp 402ndash410 2015

[8] M Niemeijer J J Staal B V Ginneken and M LoogldquoDRIVE digital retinal images for vesselextractionrdquo 2004httpwwwisiuunlResearchDatabasesDRIVE

[9] R V Kalviainen and H Uusitalo ldquoDIARETDB1 diabeticretinopathy database and evaluation protocolrdquo in Proceedingsof the 11th Conference on Medical Image Understanding andAnalysis (MIUA rsquo07) Warwick UK September 2007

[10] httpwwwadcisnetenDownload-Third-PartyMessidorhtmlindexen

[11] M Niemeijer B van Ginneken M J Cree et al ldquoRetinopathyonline challenge automatic detection of microaneurysms indigital color fundus photographsrdquo IEEE Transactions on Med-ical Imaging vol 29 no 1 pp 185ndash195 2009

[12] httpwwwadcisnetenDownload-Third-PartyE-Ophthahtml

[13] A Hoover V Kouznetsova andM Goldbaum ldquoLocating bloodvessels in retinal images by piecewise threshold probing of amatched filter responserdquo IEEETransactions onMedical Imagingvol 19 no 3 pp 203ndash210 2000

[14] S Sharma A Oliver-Fernandez W Liu P Buchholz and JWalt ldquoThe impact of diabetic retinopathy on health-relatedquality of liferdquo Current Opinion in Ophthalmology vol 16 no3 pp 155ndash159 2005

[15] G Lemaıtre M Rastgoo J Massich S Sankar F Meriaudeauand D Sidibe Classification of SD-OCT Volumes with LBPApplication to DME Detection 2015

[16] L Giancardo F Meriaudeau T P Karnowski et al ldquoTexturelessmacula swelling detection withmultiple retinal fundus imagesrdquoIEEE Transactions on Biomedical Engineering vol 58 no 3 pp795ndash799 2011

[17] V Gupta H A Al-Dhibi and J F Arevalo ldquoRetinal imaging inuveitisrdquo Saudi Journal of Ophthalmology vol 28 no 2 pp 95ndash103 2014

[18] M R KMookiah U R Acharya H Fujita et al ldquoApplication ofdifferent imaging modalities for diagnosis of Diabetic MacularEdema a reviewrdquo Computers in Biology and Medicine vol 66pp 295ndash315 2015

[19] L Reznicek S Dabov B Kayat et al ldquoScanning laser lsquoenfacersquo retinal imaging of epiretinal membranesrdquo Saudi Journal ofOphthalmology vol 28 no 2 pp 134ndash138 2014

[20] I Nakahara S Tsuruoka H Takase H Kawanaka F Okuyamaand H Matsubara ldquoExtraction of disease area from retinaloptical coherence tomography images using three dimensionalregional statisticsrdquo Procedia Computer Science vol 22 pp 893ndash901 2013

[21] F Aptel P Denis F Rouberol and C Thivolet ldquoScreeningof diabetic retinopathy effect of field number and mydriasison sensitivity and specificity of digital fundus photographyrdquoDiabetes amp Metabolism vol 34 no 3 pp 290ndash293 2008

[22] P Kahai K R Namuduri and H Thompson ldquoA decision sup-port framework for automated screening of diabetic retinopa-thyrdquo International Journal of Biomedical Imaging vol 2006Article ID 45806 8 pages 2006

[23] DUsherMDumskyjMHimaga T HWilliamson S Nusseyand J Boyce ldquoAutomated detection of diabetic retinopathy indigital retinal images a tool for diabetic retinopathy screeningrdquoDiabetic Medicine vol 21 no 1 pp 84ndash90 2004

[24] G G Gardner D Keating T H Williamson and A TElliott ldquoAutomatic detection of diabetic retinopathy using anartificial neural network a screening toolrdquo British Journal ofOphthalmology vol 80 no 11 pp 940ndash944 1996

[25] A W Reza and C Eswaran ldquoA decision support system forautomatic screening of non-proliferative diabetic retinopathyrdquoJournal of Medical Systems vol 35 no 1 pp 17ndash24 2011

[26] G S Annie Grace Vimala and S Kajamohideen ldquoDiagnosis ofdiabetic retinopathy by extracting blood vessels and exudatesusing retinal color fundus imagesrdquo WSEAS Transactions onBiology amp Biomedicine vol 11 pp 20ndash28 2014

[27] B Dupas T Walter A Erginay et al ldquoEvaluation of automatedfundus photograph analysis algorithms for detecting microa-neurysms haemorrhages and exudates and of a computer-assisted diagnostic system for grading diabetic retinopathyrdquoDiabetes amp Metabolism vol 36 no 3 pp 213ndash220 2010

Scientifica 17

[28] C Bhatia D Bhatt M Choudhary H Samant and P TaleleldquoA fast supervised retinal blood vessel segmentation usingdigital fundus imagingrdquo International Journal of Innovations ampAdvancement in Computer Science vol 4 no 6 pp 47ndash51 2015

[29] D D Rathod R R Manza Y M Rajput M B Patwari MSaswade and N Deshpande ldquoLocalization of optic disc andmacula using multilevel 2-D wavelet decomposition based onhaar wavelet transformrdquo International Journal of EngineeringResearch amp Technology IJERT vol 3 no 7 2014

[30] G B Sekar and P Nagarajan ldquoLocalisation of optic disc in fun-dus images by using clustering and histogram techniquesrdquo inProceedings of the IEEE International Conference on ComputingElectronics and Electrical Technologies (ICCEET rsquo12) pp 584ndash589 Kumaracoil India March 2012

[31] S Lu and J H Lim ldquoAutomatic optic disc detection from retinalimages by a line operatorrdquo IEEE Transactions on BiomedicalEngineering vol 58 no 1 pp 88ndash94 2011

[32] U R Acharya C K Chua L C Min E Y Ng M MMushrif and A Laude ldquoApplication of intuitionistic fuzzyhiston segmentation for the automated detection of optic discin digital fundus imagesrdquo in Proceedings of the In IEEE-EMBSInternational Conference on Biomedical and Health Informatics(BHI rsquo12) pp 444ndash447 Hong Kong January 2012

[33] E Trucco L Ballerini D Relan et al ldquoNovel VAMPIREalgorithms for quantitative analysis of the retinal vasculaturerdquoin Proceedings of the 4th ISSNIP-IEEE Biosignals and BioroboticsConference Biosignals and Robotics for Better and Safer Living(BRC rsquo13) pp 1ndash4 Rio de Janerio Brazil February 2013

[34] E Trucco A Ruggeri T Karnowski et al ldquoValidating retinalfundus image analysis algorithms issues and a proposalrdquoInvestigative Ophthalmology amp Visual Science vol 54 no 5 pp3546ndash3559 2013

[35] M R K Mookiah U R Acharya C K Chua C M Lim EY K Ng and A Laude ldquoComputer-aided diagnosis of diabeticretinopathy a reviewrdquo Computers in Biology and Medicine vol43 no 12 pp 2136ndash2155 2013

[36] C Sinthanayothin V Kongbunkiat S Phoojaruenchanachaiand A Singalavanija ldquoAutomated screening system for diabeticretinopathyrdquo in Proceedings of the 3rd International Symposiumon Image and Signal Processing and Analysis (ISPA rsquo03) vol 2pp 915ndash920 Rome Italy 2003

[37] A Osareh M Mirmehdi B Thomas and R Markham ldquoClas-sification and localisation of diabetic-related eye diseaserdquo inComputer Vision ECCV 7th European Conference on ComputerVision Copenhagen Denmark May 28ndash31 2002 ProceedingsPart IV vol 2353 of Lecture Notes in Computer Science pp 502ndash516 Springer Berlin Germany 2002

[38] G Quellec S R Russell and M D Abramoff ldquoOptimal filterframework for automated instantaneous detection of lesions inretinal imagesrdquo IEEE Transactions on Medical Imaging vol 30no 2 pp 523ndash533 2011

[39] H Fujita Y Uchiyama T Nakagawa et al ldquoComputer-aideddiagnosis the emerging of three CAD systems induced byJapanese health care needsrdquo Computer Methods and Programsin Biomedicine vol 92 no 3 pp 238ndash248 2008

[40] C I Sanchez M Niemeijer I Isgum et al ldquoContextualcomputer-aided detection improving bright lesion detection inretinal images and coronary calcification identification in CTscansrdquoMedical Image Analysis vol 16 no 1 pp 50ndash62 2012

[41] K Ram and J Sivaswamy ldquoMulti-space clustering for segmen-tation of exudates in retinal color photographsrdquo in Proceedingsof the Annual International Conference of the IEEE Engineering

in Medicine and Biology Society pp 1437ndash1440 MinneapolisMinn USA September 2009

[42] I Soares M Castelo-Branco and A M G Pinheiro ldquoExudatesdynamic detection in retinal fundus images based on the noisemap distributionrdquo in Proceedings of the 19th IEEE European Sig-nal Processing Conference (EUSIPCO rsquo11) pp 46ndash50 BarcelonaSpain September 2011

[43] C Jayakumari and T Santhanam ldquoAn intelligent approach todetect hard and soft exudates using echo state neural networkrdquoInformation Technology Journal vol 7 no 2 pp 386ndash395 2008

[44] A G Karegowda S Bhattacharyya M A Jayaram and AS Manjunath ldquoExudates detection in retinal images usingKNNFP andWKNNFP classifiersrdquo Artificial Intelligent Systemsand Machine Learning vol 3 no 7 pp 419ndash425 2011

[45] F Amel M Mohammed and B Abdelhafid ldquoImprovementof the hard exudates detection method used for computer-aided diagnosis of diabetic retinopathyrdquo International Journalof Image Graphics and Signal Processing vol 4 no 4 pp 19ndash272012

[46] P M Rokade and R R Manza ldquoAutomatic detection of hardexudates in retinal images using haar wavelet transformrdquo Eyevol 4 no 5 pp 402ndash410 2015

[47] T Jaya J Dheeba and N A Singh ldquoDetection of hard exudatesin colour fundus images using fuzzy support vector machine-based expert systemrdquo Journal of Digital Imaging vol 28 no 6pp 761ndash768 2015

[48] A Z Rozlan H Hashim S F Syed Adnan C A Hong andM Mahyudin ldquoA proposed diabetic retinopathy classificationalgorithm with statistical inference of exudates detectionrdquoin Proceedings of the International Conference on ElectricalElectronics and System Engineering (ICEESE rsquo13) pp 90ndash95IEEE Kuala Lumpur Malaysia December 2013

[49] K Soman andD Ravi ldquoDetection of exudates in human fundusimage with a comparative study on methods for the optic diskdetectionrdquo in Proceedings of the IEEE International Conferenceon Information Communication and Embedded Systems (ICICESrsquo14) pp 1ndash5 Chennai India February 2014

[50] R Annunziata A Garzelli L Ballerini A Mecocci and ETrucco ldquoLeveraging multiscale hessian-based enhancementwith a novel exudate inpainting technique for retinal vessel seg-mentationrdquo IEEE Journal of Biomedical and Health Informaticsvol 20 no 4 pp 1129ndash1138 2016

[51] M J J P Van Grinsven A Chakravarty J Sivaswamy TTheelen B Van Ginneken and C I Sanchez ldquoA bag ofwords approach for discriminating between retinal imagescontaining exudates or drusenrdquo in Proceedings of the IEEE 10thInternational Symposium on Biomedical Imaging from Nano toMacro (ISBI rsquo13) pp 1444ndash1447 San Francisco Calif USA April2013

[52] J Kaur and D Mittal ldquoSegmentation and measurement ofexudates in fundus images of the retina for detection ofretinal diseaserdquo Journal of Biomedical Engineering and MedicalImaging vol 2 no 1 pp 27ndash38 2015

[53] A Sopharak B Uyyanonvara and S Barman ldquoAutomaticmicroaneurysm detection from non-dilated diabetic retinopa-thy retinal images using mathematical morphology methodsrdquoIAENG International Journal of Computer Science vol 38 no 3pp 295ndash301 2011

[54] N V Krishna N V Reddy M V Ramana and E P KumarldquoThe communal system for early detection microaneurysm anddiabetic retinopathy grading through color fundus imagesrdquo

18 Scientifica

International Journal of Scientific Engineering and Technologyvol 2 pp 228ndash232 2013

[55] R Roy S Aruchamy and P Bhattacharjee ldquoDetection of retinalmicroaneurysms using fractal analysis and feature extractiontechniquerdquo in Proceedings of the 2nd International Conferenceon Communication and Signal Processing (ICCSP rsquo13) pp 469ndash474 Melmaruvathur India April 2013

[56] K Adal S Ali D Sidibe T Karnowski E Chaum and FMeriaudeau ldquoAutomated detection of microaneurysms usingrobust blob descriptorsrdquo in Medical Imaging 2013 Computer-Aided Diagnosis vol 8670 of Proceedings of SPIE InternationalSociety for Optics and Photonics March 2013

[57] N S Datta H S Dutta M De and S Mondal ldquoAn effectiveapproach image quality enhancement for microaneurysmsdetection of non-dilated retinal fundus imagerdquo Procedia Tech-nology vol 10 pp 731ndash737 2013

[58] L Giancardo F Meriaudeau T P Karnowski K W Tobin YLi and E Chaum ldquoMicroaneurysms detection with the radoncliff operator in retinal fundus imagesrdquo inMedical Imaging 2010Image Processing vol 7623 of Proceedings of SPIE San DiegoCalif USA February 2010

[59] S Ding and W Ma ldquoAn accurate approach for microaneurysmdetection in digital fundus imagesrdquo in Proceedings of the 22ndInternational Conference on Pattern Recognition (ICPR rsquo14) pp1846ndash1851 IEEE Stockholm Sweden August 2014

[60] RMRangayyanX Zhu F J Ayres andA L Ells ldquoDetection ofthe optic nerve head in fundus images of the retina with gaborfilters and phase portrait analysisrdquo Journal of Digital Imagingvol 23 no 4 pp 438ndash453 2010

[61] FOloumi RMRangayyan andA L Ells ldquoParabolicmodelingof the major temporal arcade in retinal fundus imagesrdquo IEEETransactions on Instrumentation and Measurement vol 61 no7 pp 1825ndash1838 2012

[62] A Giachetti K S Chin E Trucco C Cobb and P J WilsonldquoMultiresolution localization and segmentation of the opticaldisc in fundus images using inpainted background and vesselinformationrdquo in Proceedings of the 18th IEEE International Con-ference on Image Processing (ICIP rsquo11) pp 2145ndash2148 BrusselsBelgium September 2011

[63] A Dehghani H A Moghaddam and M-S Moin ldquoOpticdisc localization in retinal images using histogram matchingrdquoEURASIP Journal on Image and Video Processing vol 2012article 19 pp 1ndash11 2012

[64] TWalter J-C Klein P Massin and A Erginay ldquoA contributionof image processing to the diagnosis of diabetic retinopathy-detection of exudates in color fundus images of the humanretinardquo IEEE Transactions on Medical Imaging vol 21 no 10pp 1236ndash1243 2002

[65] A Dehghani M-S Moin and M Saghafi ldquoLocalization of theoptic disc center in retinal images based on the Harris cornerdetectorrdquo Biomedical Engineering Letters vol 2 no 3 pp 198ndash206 2012

[66] JMMolina and E J Carmona ldquoLocalization and segmentationof the optic nerve head in eye fundus images using pyramidrepresentation and genetic algorithmsrdquo in Foundations onNatural and Artificial Computation vol 6686 of Lecture Notesin Computer Science pp 431ndash440 Springer Berlin Germany2011

[67] A Giachetti L Ballerini E Trucco and P J Wilson ldquoThe useof radial symmetry to localize retinal landmarksrdquoComputerizedMedical Imaging andGraphics vol 37 no 5-6 pp 369ndash376 2013

[68] L Xiong and H Li ldquoAn approach to locate optic disc in reti-nal images with pathological changesrdquo Computerized MedicalImaging and Graphics vol 47 pp 40ndash50 2016

[69] J H Tan and U Rajendra Acharya ldquoActive spline model ashape based modelmdashinteractive segmentationrdquo Digital SignalProcessing A Review Journal vol 35 pp 64ndash74 2014

[70] Y Hatanaka T Nakagawa Y Hayashi et al ldquoImprovement ofautomatic hemorrhage detectionmethods using brightness cor-rection on fundus imagesrdquo inMedical Imaging 2008 Computer-Aided Diagnosis 69153E vol 6915 of Proceedings of SPIEInternational Society for Optics and Photonics March 2008

[71] L Tang M Niemeijer J M Reinhardt M K Garvin andM DAbramoff ldquoSplat feature classificationwith application to retinalhemorrhage detection in fundus imagesrdquo IEEE Transactions onMedical Imaging vol 32 no 2 pp 364ndash375 2013

[72] M D Saleh and C Eswaran ldquoAn automated decision-supportsystem for non-proliferative diabetic retinopathy disease basedon MAs and HAs detectionrdquo Computer Methods and Programsin Biomedicine vol 108 no 1 pp 186ndash196 2012

[73] J Lachure A V Deorankar S Lachure S Gupta and RJadhav ldquoDiabetic Retinopathy using morphological operationsand machine learningrdquo in Proceedings of the 5th IEEE Interna-tional Advance Computing Conference (IACC rsquo15) pp 617ndash622Banglore India June 2015

[74] L R Sudha and S Thirupurasundari ldquoAnalysis and detectionof haemorrhages and exudates in retinal imagesrdquo InternationalJournal of Scientific and Research Publications vol 4 no 3 2014

[75] A Sharma M K Dutta A Singh M Parthasarathi and CM Travieso ldquoDynamic thresholding technique for detectionof hemorrhages in retinal imagesrdquo in Proceedings of the 7thInternational Conference on Contemporary Computing (IC3 rsquo14)pp 113ndash116 IEEE Noida India August 2014

[76] M U Akram S Khalid A Tariq S A Khan and F AzamldquoDetection and classification of retinal lesions for grading ofdiabetic retinopathyrdquo Computers in Biology and Medicine vol45 no 1 pp 161ndash171 2014

[77] M N Ashraf Z Habib and M Hussain ldquoTexture featureanalysis of digital fundus images for early detection of diabeticretinopathyrdquo in Proceedings of the 11th IEEE InternationalConference on Computer Graphics Imaging and Visualization(CGIV rsquo14) pp 57ndash62 Singapore August 2014

[78] M M Fraz P Remagnino A Hoppe et al ldquoAn ensembleclassification-based approach applied to retinal blood vesselsegmentationrdquo IEEE Transactions on Biomedical Engineeringvol 59 no 9 pp 2538ndash2548 2012

[79] U T V Nguyen A Bhuiyan L A F Park and K Ramamoha-narao ldquoAn effective retinal blood vessel segmentation methodusing multi-scale line detectionrdquo Pattern Recognition vol 46no 3 pp 703ndash715 2013

[80] M M Fraz P Remagnino A Hoppe and S A BarmanldquoRetinal image analysis aimed at extraction of vascular structureusing linear discriminant classifierrdquo in Proceedings of the Inter-national Conference on ComputerMedical Applications (ICCMArsquo13) pp 1ndash6 Sousse Tunisia January 2013

[81] X Yin B W-H Ng J He Y Zhang and D Abbott ldquoAccu-rate image analysis of the retina using hessian matrix andbinarisation of thresholded entropy with application of texturemappingrdquo PLoS ONE vol 9 no 4 Article ID e95943 2014

[82] R Vega E Guevara L Falcon G Sanchez-Ante and HSossa ldquoBlood vessel segmentation in retinal images using latticeneural networksrdquo in Advances in Artificial Intelligence and Its

Scientifica 19

Applications 12thMexican International Conference onArtificialIntelligence MICAI 2013 Mexico City Mexico November 24ndash302013 Proceedings Part I vol 8265 of Lecture Notes in ComputerScience pp 532ndash544 Springer Berlin Germany 2013

[83] D Marın A Aquino M E Gegundez-Arias and J M BravoldquoA new supervised method for blood vessel segmentation inretinal images by using gray-level andmoment invariants-basedfeaturesrdquo IEEE Transactions on Medical Imaging vol 30 no 1pp 146ndash158 2011

[84] Y Hou ldquoAutomatic segmentation of retinal blood vessels basedon improved multiscale line detectionrdquo Journal of ComputingScience and Engineering vol 8 no 2 pp 119ndash128 2014

[85] F Shami H Seyedarabi and A Aghagolzadeh ldquoBetter detec-tion of retinal abnormalities by accurate detection of bloodvessels in retinardquo in Proceedings of the 22nd Iranian Conferenceon Electrical Engineering (ICEE rsquo14) pp 1493ndash1496 Tehran IranMay 2014

[86] S S Basha and K S Prasad ldquoAutomatic detection of hardexudates in diabetic retinopathy using morphological seg-mentation and fuzzy logicrdquo International Journal of ComputerScience and Network Security vol 8 no 12 pp 211ndash218 2008

[87] M Garcıa C I Sanchez M I Lopez D Abasolo andR Hornero ldquoNeural network based detection of hard exu-dates in retinal imagesrdquo Computer Methods and Programs inBiomedicine vol 93 no 1 pp 9ndash19 2009

[88] A Somasundaram and J Prabhu ldquoDetection of exudates forthe diagnosis of diabetic retinopathyrdquo International Journal ofInnovation and Applied Studies vol 3 no 1 pp 116ndash120 2013

[89] A Kaur and P Kaur ldquoA comparative study of various exudatesegmentation techniques for diagnosis of diabetic retinopathyrdquoInternational Journal of Current Engineering and Technologyvol 46 no 1 pp 142ndash146 2016

[90] V Saravanan B Venkatalakshmi and S M Farhana ldquoDesignand development of pervasive classifier for diabetic retinopa-thyrdquo in Proceedings of the IEEE Information amp CommunicationTechnologies (ICT rsquo13) pp 231ndash235 JeJu Island South KoreaApril 2013

[91] TWalter J-C Klein P Massin and A Erginay ldquoA contributionof image processing to the diagnosis of diabetic retinopathymdashdetection of exudates in color fundus images of the humanretinardquo IEEE Transactions on Medical Imaging vol 21 no 10pp 1236ndash1243 2002

[92] S Ali D Sidibe KM Adal et al ldquoStatistical atlas based exudatesegmentationrdquo Computerized Medical Imaging and Graphicsvol 37 no 5-6 pp 358ndash368 2013

[93] L Giancardo F Meriaudeau T P Karnowski et al ldquoExudate-based diabeticmacular edema detection in fundus images usingpublicly available datasetsrdquoMedical Image Analysis vol 16 no1 pp 216ndash226 2012

[94] N Bhagat R A Grigorian A Tutela and M A ZarbinldquoDiabetic macular edema pathogenesis and treatmentrdquo Surveyof Ophthalmology vol 54 no 1 pp 1ndash32 2009

[95] J Sebag and E A Balazs ldquoPathogenesis of cystoid macularedema an anatomic consideration of vitreoretinal adhesionsrdquoSurvey of Ophthalmology vol 28 supplement 2 pp 493ndash4981984

[96] B Y Kim S D Smith and P K Kaiser ldquoOptical coherencetomographic patterns of diabetic macular edemardquo AmericanJournal of Ophthalmology vol 142 no 3 pp 405ndash412e1 2006

[97] D M Brown and C D Regillo ldquoAnti-VEGF agents in thetreatment of neovascular age-related macular degeneration

applying clinical trial results to the treatment of everydaypatientsrdquo American Journal of Ophthalmology vol 144 no 4pp 627ndash637e2 2007

[98] A C Ho I U Scott S J Kim et al ldquoAntindashvascular endothelialgrowth factor pharmacotherapy for diabetic macular edema areport by the American academy of ophthalmologyrdquo Ophthal-mology vol 119 no 10 pp 2179ndash2188 2012

[99] P Romero-Aroca ldquoCurrent status in diabetic macular edematreatmentsrdquoWorld Journal of Diabetes vol 4 no 5 pp 165ndash1692013

[100] S Vujosevic F Martini E Convento et al ldquoSubthresholdlaser therapy for diabetic macular edema metabolic and safetyissuesrdquo Current Medicinal Chemistry vol 20 no 26 pp 3267ndash3271 2013

[101] M U Akram A Tariq S A Khan andM Y Javed ldquoAutomateddetection of exudates and macula for grading of diabetic mac-ular edemardquo Computer Methods and Programs in Biomedicinevol 114 no 2 pp 141ndash152 2014

[102] J B Raja and C G Ravichandran ldquoAutomatic localization offovea in retinal images based onmathematical morphology andanatomic structuresrdquo International Journal of Engineering andTechnology vol 6 no 5 pp 2171ndash2183 2014

[103] T Walter and J C Klein ldquoAutomatic detection of microa-neurysms in color fundus images of the human retina bymeansof the bounding box closingrdquo inMedical Data Analysis pp 210ndash220 Springer Berlin Germany 2002

[104] B Zhang X Wu J You Q Li and F Karray ldquoDetectionof microaneurysms using multi-scale correlation coefficientsrdquoPattern Recognition vol 43 no 6 pp 2237ndash2248 2010

[105] P J Navarro D Alonso and K Stathis ldquoAutomatic detection ofmicroaneurysms in diabetic retinopathy fundus images usingthe Llowastalowastb color spacerdquo Journal of the Optical Society of AmericaA vol 33 no 1 pp 74ndash83 2016

[106] H Narasimha-Iyer A Can B Roysam et al ldquoRobust detectionand classification of longitudinal changes in color retinal fundusimages for monitoring diabetic retinopathyrdquo IEEE Transactionson Biomedical Engineering vol 53 no 6 pp 1084ndash1098 2006

[107] K K Delibasis A I Kechriniotis C Tsonos and N AssimakisldquoAutomaticmodel-based tracing algorithm for vessel segmenta-tion anddiameter estimationrdquoComputerMethods andProgramsin Biomedicine vol 100 no 2 pp 108ndash122 2010

[108] M D Saleh and C Eswaran ldquoAn efficient algorithm for retinalblood vessel segmentation using h-maxima transform andmultilevel thresholdingrdquo Computer Methods in Biomechanicsand Biomedical Engineering vol 15 no 5 pp 517ndash525 2012

[109] N P Singh and R Srivastava ldquoRetinal blood vessels seg-mentation by using Gumbel probability distribution functionbased matched filterrdquo Computer Methods and Programs inBiomedicine vol 129 pp 40ndash50 2016

[110] K E Pumama M H F Wilkinson A G Veidhuizen et alldquoBranches filtering approach for max-treerdquo in Proceedings of the2nd International Conference on Computer Vision Theory andApplications (VISAPP rsquo07) vol 1 pp 328ndash332 Barcelona SpainMarch 2007

[111] S Supot CThanapong P Chuchart and S Manas ldquoAutomaticsegmentation of blood vessels in retinal image based on fuzzyK-median clusteringrdquo in Proceedings of the IEEE InternationalConference on Integration Technology (ICIT rsquo07) pp 584ndash588IEEE Shenzhen China March 2007

[112] D SatyarthiM R Kumar and S Dandapat ldquoGaussian intensitydistribution modelling of blood vessels in fundus imagesrdquo in

20 Scientifica

Proceedings of the Annual IEEE India Conference (INDICONrsquo05) pp 228ndash232 Chennai India December 2005

[113] A Z Rozlan N S Mohd Ali and H Hashim ldquoGUI systemfor enhancing blood vessels segmentation in digital fundusimagesrdquo in Proceedings of the IEEEControl and SystemGraduateResearch Colloquium (ICSGRC rsquo12) pp 55ndash59 Shah AlamMalaysia July 2012

[114] F-L Yi and W-H Xu ldquoSegmentation of blood vessels in colorfundus images based on optimal multi-threshold methodrdquo inProceedings of the International Conference onMachine Learningand Cybernetics (ICMLC rsquo12) vol 2 pp 725ndash728 IEEE XirsquoanChina July 2012

[115] J Sivakamasundari G Kavitha VNatarajan and S Ramakrish-nan ldquoProposal of a Content Based retinal Image Retrieval sys-tem using Kirsch template based edge detectionrdquo in Proceedingsof the IEEE International Conference on Informatics Electronicsamp Vision (ICIEV rsquo14) pp 1ndash5 Dhaka Bangladesh May 2014

[116] J Amin and A Zafar ldquoA survey content based image retrievalrdquoThe International Journal of Advanced Networking and Applica-tions (IJANA) vol 6 pp 2076ndash2083 2014

Submit your manuscripts athttpwwwhindawicom

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


MEDIATORSINFLAMMATION

of


Behavioural Neurology

EndocrinologyInternational Journal of



Disease Markers


BioMed Research International

OncologyJournal of



Oxidative Medicine and Cellular Longevity


PPAR Research

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

ObesityJournal of



Computational and Mathematical Methods in Medicine

OphthalmologyJournal of


Diabetes ResearchJournal of



Research and TreatmentAIDS


Gastroenterology Research and Practice


Parkinsonrsquos Disease

Evidence-Based Complementary and Alternative Medicine

Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom

2 Scientifica

MAs HMsEXs

(a)

Abnormal growth of vessels

(b)

Figure 1 Stages of diabetic retinopathy (a) Signs of NPDR (b) Signs of PDR

2 Publicly Available Databases

Images of retina are taken by a device called fundus cameraRetinal fundus images (RFI) is the name given to theseimages This camera takes images of the internal surface ofretina posterior pole macula optic disc and blood vesselsImage acquisition is a leading step for medical diagnosisFor research on medical fundus images widely accessibleresources are available as shown in Table 1 Some benchmarkdatabases are openly available for the assessment of algo-rithms introduced for the computerized screening and anal-ysis of DR The purpose of databases is to check the strengthof automatic screening of DR and then compare the resultswith current techniques Seven datasets are available openlyincluding DRIVE STARE DIARETDB e-ophtha HEI-MED Retinopathy Online Challenge (ROC) and Messidor

3 Performance Measures

For the diagnosis of DR in its early stage the retinal imagecaptured using some fundus camera needs to undergo prepr-ocessing before applying further algorithms of image proces-sing Many preprocessing techniques such as contrast adjust-ment average filtering adaptive histogram equalizationhomomorphic and median filtering are applied on retinalimages in the gold standard database After an algorithm isapplied to a retinal image mean square error (MSE) andpeak signal to noise ratio (PSNR) are further calculated toanalyze the performance of an algorithm PSNR is mostlydenoted regarding logarithmic decibel value A higher valueof PSNR means that the processed image is of higher andbetter quality than the actual image In medical treatmentthe medical contribution data is frequently divided into twotypes data where the disease is present and data where thedisease is not diagnosedThe level of correctness of treatmentis reviewed by the sensitivity and specificity measures Inmedical research fundus images which are common in DRare calculated via sensitivity and specificity of each imageThe higher sensitivity and specificity values enhance thetreatment True positive (TP) denotes the total number oflesion pixels and true negative (TN) denotes the nonlesionpixels False positive (FP) denotes the number of nonlesionpixels that are detected wrongly by the algorithm Likewisefalse negative (FN) indicates the number of lesion pixels

that are not identified by the algorithm Table 2 showsperformance metrics

4 Methods for Detection of DiabeticRetinopathy

DR is a main reason of blindness DR development is at dif-ferent rates in different persons because of the two importantvision pressuring difficulties diabetic macular edema (DME)and proliferative retinopathy That is why there is a hugedemand for the latest technologies andmethodologies to ana-lyze DR efficiently and correctly in its early stage Nowadaysthe research community has givenmany techniques shown inFigure 2 for early detection of DR that is conferred here

5 Screening of Diabetic Retinopathy

Eye illnesses for example DR and DME are the most widelyrecognized reasons for irreversible vision loss in people withdiabetes Just in the United States alone health care andrelated costs identified with eye maladies are assessed atnearly $500M Besides the prevalent cases of DR are reliedupon to become exponentially influencing more than 300Mindividuals around the world by 2025 Early discovery andtreatment of DR and DME play a major role in prevent-ing adverse effects such as blindness Optical CoherenceTomography (OCT) imaging is ideal for uncovering DME onaccount of its extended axial resolution and extended retinalscan coverage spatial domain optical coherence tomography(SD-OCT) revealed DME-related changes of the photore-ceptors external restricting film Bruchrsquos layer and retinalpigment epithelium layer Moreover SD-OCT perceivedrelated epiretinal layers changes in the retinal decay andvitreoretinal interface The issues of automatic classificationof SD-OCT information for automatically recognizable proofof patients with DME versus ordinary subjects are tendedas well The proposed technique depends on Local BinaryPattern (LBP) features to portray the texture of OCT picturesand contrast diverse LBP features extraction approaches withthe process of single signature for the entire OCT volumeTest results with two datasets of separately 32 and 30 OCTvolumes demonstrate that regardless to utilizing low or highstate representations features obtained from LBP texture

Scientifica 3

Table1Databases

forfun

dusimages

Nam

eIm

agea

cquisition

Num

bero

fimages

Resolutio

nUses

DRIVE[8]

3-CC

Dcameraw

ith45-fo

ldview

20colorfun

dustestin

gim

ages

20colorfun

dustrainingim

ages

768times584

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

Image-Re

t(DIARE

TB0

DIARE

TB1)[9]

50-fo

ldview

DIARE

TB0total130

images

inwhich

20im

ages

aren

ormaland110

with

DR

DIARE

TB1total89im

agesfive

images

aren

ormaland84

images

with

DR

1500times1152

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

Messid

or[10]

3CCD

cameraa

t45-fold

view

1200

images


2304times1536

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

Retin

opathy

OnlineC

hallenge

[11]

Cano

nCR

5-45-N

Mcameraa

t45-fo

ldview

100digitalfun

dusimages


1389times1383

Microaneurysm

sdetectio

n

e-op

htha

EX[12]

OPH

DIATcopy

Tele-m

edicalnetwork

Itcontains

47im

ages

with

exud

ates

and

35im

ages

with

nolesio

n2048times1360

Exud

ates

detection

e-op

htha

MA[12]

OPH

DIATcopy

Tele-m

edicalnetwork

Itcontains

148im

ages

with

microaneurysm

sorsmallh

emorrhages

and233im

ages

with

nolesio

n2048times1360

Microaneurysm

sdetectio

n

STARE

[13]

TopconTR

V50

fund

uscameraa

t35-fo

ldview

400im

ages

605times700

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

HEI-M

ED[7]

Zeiss

Visucam

PROfund

uscamera

at45-fo

ldview

169im

ages

inwhich

115im

ages

are

abno

rmaland54

images

areh

ealth

y2196times1958

Exud

ates

detection

4 Scientifica







TPR = TPP=



FPFP + TN





SPC = TNN=





minusinfin

infin




Retinal imaging


Nonproliferative

Proliferative

Microaneurysm

Macular edema

Exudates

Hemorrhages




6 Retinal Imaging


Scientifica 5






Optic disc




6 Scientifica

Table3CA

Dmetho

dsford

iagn

osisof

DR

Algorith

mIm

agep

rocessing

techniqu

esDatabase

Color

space

Sensitivity

Specificity

AccuracyAUC

Lesio

nsdetection

Apteletal[21]

Sing

le-field

nonm

ydria

tic

single-field

mydria

tic

three-field

nonm

ydria

tic

three-field

mydria

tic

79patie

nts(158eyes)

Grayscale

7790

92

97

9998

97

98

082090090095

Both

NPD

RandPD

R

Kahaietal[22]

Decision

supp

ort

syste

m(D

SS)

143im

ages

Louisia

naState

University

Eye

Center

Grayscale

100

67

mdashNPD

R

Usher

etal[23]

Neuraln

etwork

1273

consecutive

patie

ntsSt

Thom

asHospital

Grayscale

948

mdashmdash

NPD

R

Gardn

eretal[24]

Back

prop

agation

neuralnetwork

200diabetic

and101n

ormal

images

ofprivate

hospita

l

Grayscale

884

835

mdashBo

thNPD

RandPD

R

Reza

andEswaran

[25]

Rulebasedcla

ssifier

STARE

Green

channel

972

100

97

NPD

R

Ann

ieGrace

Vimalaa

ndKa

jamoh

ideen[26]

Cost-e

ffective

compu

ter-aideddiagno

stic

syste

m

Privatee

yeHospital

HSV

916

905

912

NPD

R

Dup

asetal[27]

Automated

fund

usph

otograph

analysis

algorithm

sMessid

orGrayscale

839

727

mdashNPD

R

Ashrafetal[77]

LocalB

inaryPatte

rn

SVM

DIARE

TB1

Green

channel

8748

8599

8615

087

NPD

R

Scientifica 7

Table4Differentm

etho

dsford

etectio

nof

optic

disc

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Accuracy

Rathod

etal[29]

Multilevel2D

wavele

tHistogram

Equalization

HRF

Green

channel

95

Sekara

ndNagarajan

[30]


cluste

ringandhisto

gram

approaches

Messid

orRG

B9958

LuandLim

[31]

Circular

transfo

rmation

lineo

perator

DIARE

TDB0

DIARE

TDB1

DRIVE

STARE

CIEL

AB

974

Acharyae

tal[32]

Otsu

Gradientvectorfl

ow(G

VF)

snakeAtanassov

Intuition

isticFu

zzyHiston

(AIFSH

)Ka

sturbaM

edicalCollege

Grayscale

100

Trucco

etal[33]

Binary

vesselmasks

ared

evelo

pedwith

inVA

MPIRE

softw

are

HRIS(H

igh-Re

solutio

nIm

age

Set)

VDIS

(VascularD

iseaseImage

Set)

Grayscale

957

921

8 Scientifica







Scientifica 9






Microaneurysms

Microaneurysms





10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE


HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Scientifica 3

Table1Databases

forfun

dusimages

Nam

eIm

agea

cquisition

Num

bero

fimages

Resolutio

nUses

DRIVE[8]

3-CC

Dcameraw

ith45-fo

ldview

20colorfun

dustestin

gim

ages

20colorfun

dustrainingim

ages

768times584

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

Image-Re

t(DIARE

TB0

DIARE

TB1)[9]

50-fo

ldview

DIARE

TB0total130

images

inwhich

20im

ages

aren

ormaland110

with

DR

DIARE

TB1total89im

agesfive

images

aren

ormaland84

images

with

DR

1500times1152

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

Messid

or[10]

3CCD

cameraa

t45-fold

view

1200

images


2304times1536

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

Retin

opathy

OnlineC

hallenge

[11]

Cano

nCR

5-45-N

Mcameraa

t45-fo

ldview

100digitalfun

dusimages


1389times1383

Microaneurysm

sdetectio

n

e-op

htha

EX[12]

OPH

DIATcopy

Tele-m

edicalnetwork

Itcontains

47im

ages

with

exud

ates

and

35im

ages

with

nolesio

n2048times1360

Exud

ates

detection

e-op

htha

MA[12]

OPH

DIATcopy

Tele-m

edicalnetwork

Itcontains

148im

ages

with

microaneurysm

sorsmallh

emorrhages

and233im

ages

with

nolesio

n2048times1360

Microaneurysm

sdetectio

n

STARE

[13]

TopconTR

V50

fund

uscameraa

t35-fo

ldview

400im

ages

605times700

Exud

ateshem

orrhages

microaneurysm

sand

abno

rmalbloo

dvessels

detection

HEI-M

ED[7]

Zeiss

Visucam

PROfund

uscamera

at45-fo

ldview

169im

ages

inwhich

115im

ages

are

abno

rmaland54

images

areh

ealth

y2196times1958

Exud

ates

detection

4 Scientifica







TPR = TPP=



FPFP + TN





SPC = TNN=





minusinfin

infin




Retinal imaging


Nonproliferative

Proliferative

Microaneurysm

Macular edema

Exudates

Hemorrhages




6 Retinal Imaging


Scientifica 5






Optic disc




6 Scientifica

Table3CA

Dmetho

dsford

iagn

osisof

DR

Algorith

mIm

agep

rocessing

techniqu

esDatabase

Color

space

Sensitivity

Specificity

AccuracyAUC

Lesio

nsdetection

Apteletal[21]

Sing

le-field

nonm

ydria

tic

single-field

mydria

tic

three-field

nonm

ydria

tic

three-field

mydria

tic

79patie

nts(158eyes)

Grayscale

7790

92

97

9998

97

98

082090090095

Both

NPD

RandPD

R

Kahaietal[22]

Decision

supp

ort

syste

m(D

SS)

143im

ages

Louisia

naState

University

Eye

Center

Grayscale

100

67

mdashNPD

R

Usher

etal[23]

Neuraln

etwork

1273

consecutive

patie

ntsSt

Thom

asHospital

Grayscale

948

mdashmdash

NPD

R

Gardn

eretal[24]

Back

prop

agation

neuralnetwork

200diabetic

and101n

ormal

images

ofprivate

hospita

l

Grayscale

884

835

mdashBo

thNPD

RandPD

R

Reza

andEswaran

[25]

Rulebasedcla

ssifier

STARE

Green

channel

972

100

97

NPD

R

Ann

ieGrace

Vimalaa

ndKa

jamoh

ideen[26]

Cost-e

ffective

compu

ter-aideddiagno

stic

syste

m

Privatee

yeHospital

HSV

916

905

912

NPD

R

Dup

asetal[27]

Automated

fund

usph

otograph

analysis

algorithm

sMessid

orGrayscale

839

727

mdashNPD

R

Ashrafetal[77]

LocalB

inaryPatte

rn

SVM

DIARE

TB1

Green

channel

8748

8599

8615

087

NPD

R

Scientifica 7

Table4Differentm

etho

dsford

etectio

nof

optic

disc

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Accuracy

Rathod

etal[29]

Multilevel2D

wavele

tHistogram

Equalization

HRF

Green

channel

95

Sekara

ndNagarajan

[30]


cluste

ringandhisto

gram

approaches

Messid

orRG

B9958

LuandLim

[31]

Circular

transfo

rmation

lineo

perator

DIARE

TDB0

DIARE

TDB1

DRIVE

STARE

CIEL

AB

974

Acharyae

tal[32]

Otsu

Gradientvectorfl

ow(G

VF)

snakeAtanassov

Intuition

isticFu

zzyHiston

(AIFSH

)Ka

sturbaM

edicalCollege

Grayscale

100

Trucco

etal[33]

Binary

vesselmasks

ared

evelo

pedwith

inVA

MPIRE

softw

are

HRIS(H

igh-Re

solutio

nIm

age

Set)

VDIS

(VascularD

iseaseImage

Set)

Grayscale

957

921

8 Scientifica







Scientifica 9






Microaneurysms

Microaneurysms





10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE


HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















4 Scientifica







TPR = TPP=



FPFP + TN





SPC = TNN=





minusinfin

infin




Retinal imaging


Nonproliferative

Proliferative

Microaneurysm

Macular edema

Exudates

Hemorrhages




6 Retinal Imaging


Scientifica 5






Optic disc




6 Scientifica

Table3CA

Dmetho

dsford

iagn

osisof

DR

Algorith

mIm

agep

rocessing

techniqu

esDatabase

Color

space

Sensitivity

Specificity

AccuracyAUC

Lesio

nsdetection

Apteletal[21]

Sing

le-field

nonm

ydria

tic

single-field

mydria

tic

three-field

nonm

ydria

tic

three-field

mydria

tic

79patie

nts(158eyes)

Grayscale

7790

92

97

9998

97

98

082090090095

Both

NPD

RandPD

R

Kahaietal[22]

Decision

supp

ort

syste

m(D

SS)

143im

ages

Louisia

naState

University

Eye

Center

Grayscale

100

67

mdashNPD

R

Usher

etal[23]

Neuraln

etwork

1273

consecutive

patie

ntsSt

Thom

asHospital

Grayscale

948

mdashmdash

NPD

R

Gardn

eretal[24]

Back

prop

agation

neuralnetwork

200diabetic

and101n

ormal

images

ofprivate

hospita

l

Grayscale

884

835

mdashBo

thNPD

RandPD

R

Reza

andEswaran

[25]

Rulebasedcla

ssifier

STARE

Green

channel

972

100

97

NPD

R

Ann

ieGrace

Vimalaa

ndKa

jamoh

ideen[26]

Cost-e

ffective

compu

ter-aideddiagno

stic

syste

m

Privatee

yeHospital

HSV

916

905

912

NPD

R

Dup

asetal[27]

Automated

fund

usph

otograph

analysis

algorithm

sMessid

orGrayscale

839

727

mdashNPD

R

Ashrafetal[77]

LocalB

inaryPatte

rn

SVM

DIARE

TB1

Green

channel

8748

8599

8615

087

NPD

R

Scientifica 7

Table4Differentm

etho

dsford

etectio

nof

optic

disc

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Accuracy

Rathod

etal[29]

Multilevel2D

wavele

tHistogram

Equalization

HRF

Green

channel

95

Sekara

ndNagarajan

[30]


cluste

ringandhisto

gram

approaches

Messid

orRG

B9958

LuandLim

[31]

Circular

transfo

rmation

lineo

perator

DIARE

TDB0

DIARE

TDB1

DRIVE

STARE

CIEL

AB

974

Acharyae

tal[32]

Otsu

Gradientvectorfl

ow(G

VF)

snakeAtanassov

Intuition

isticFu

zzyHiston

(AIFSH

)Ka

sturbaM

edicalCollege

Grayscale

100

Trucco

etal[33]

Binary

vesselmasks

ared

evelo

pedwith

inVA

MPIRE

softw

are

HRIS(H

igh-Re

solutio

nIm

age

Set)

VDIS

(VascularD

iseaseImage

Set)

Grayscale

957

921

8 Scientifica







Scientifica 9






Microaneurysms

Microaneurysms





10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE


HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Scientifica 5






Optic disc




6 Scientifica

Table3CA

Dmetho

dsford

iagn

osisof

DR

Algorith

mIm

agep

rocessing

techniqu

esDatabase

Color

space

Sensitivity

Specificity

AccuracyAUC

Lesio

nsdetection

Apteletal[21]

Sing

le-field

nonm

ydria

tic

single-field

mydria

tic

three-field

nonm

ydria

tic

three-field

mydria

tic

79patie

nts(158eyes)

Grayscale

7790

92

97

9998

97

98

082090090095

Both

NPD

RandPD

R

Kahaietal[22]

Decision

supp

ort

syste

m(D

SS)

143im

ages

Louisia

naState

University

Eye

Center

Grayscale

100

67

mdashNPD

R

Usher

etal[23]

Neuraln

etwork

1273

consecutive

patie

ntsSt

Thom

asHospital

Grayscale

948

mdashmdash

NPD

R

Gardn

eretal[24]

Back

prop

agation

neuralnetwork

200diabetic

and101n

ormal

images

ofprivate

hospita

l

Grayscale

884

835

mdashBo

thNPD

RandPD

R

Reza

andEswaran

[25]

Rulebasedcla

ssifier

STARE

Green

channel

972

100

97

NPD

R

Ann

ieGrace

Vimalaa

ndKa

jamoh

ideen[26]

Cost-e

ffective

compu

ter-aideddiagno

stic

syste

m

Privatee

yeHospital

HSV

916

905

912

NPD

R

Dup

asetal[27]

Automated

fund

usph

otograph

analysis

algorithm

sMessid

orGrayscale

839

727

mdashNPD

R

Ashrafetal[77]

LocalB

inaryPatte

rn

SVM

DIARE

TB1

Green

channel

8748

8599

8615

087

NPD

R

Scientifica 7

Table4Differentm

etho

dsford

etectio

nof

optic

disc

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Accuracy

Rathod

etal[29]

Multilevel2D

wavele

tHistogram

Equalization

HRF

Green

channel

95

Sekara

ndNagarajan

[30]


cluste

ringandhisto

gram

approaches

Messid

orRG

B9958

LuandLim

[31]

Circular

transfo

rmation

lineo

perator

DIARE

TDB0

DIARE

TDB1

DRIVE

STARE

CIEL

AB

974

Acharyae

tal[32]

Otsu

Gradientvectorfl

ow(G

VF)

snakeAtanassov

Intuition

isticFu

zzyHiston

(AIFSH

)Ka

sturbaM

edicalCollege

Grayscale

100

Trucco

etal[33]

Binary

vesselmasks

ared

evelo

pedwith

inVA

MPIRE

softw

are

HRIS(H

igh-Re

solutio

nIm

age

Set)

VDIS

(VascularD

iseaseImage

Set)

Grayscale

957

921

8 Scientifica







Scientifica 9






Microaneurysms

Microaneurysms





10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE


HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















6 Scientifica

Table3CA

Dmetho

dsford

iagn

osisof

DR

Algorith

mIm

agep

rocessing

techniqu

esDatabase

Color

space

Sensitivity

Specificity

AccuracyAUC

Lesio

nsdetection

Apteletal[21]

Sing

le-field

nonm

ydria

tic

single-field

mydria

tic

three-field

nonm

ydria

tic

three-field

mydria

tic

79patie

nts(158eyes)

Grayscale

7790

92

97

9998

97

98

082090090095

Both

NPD

RandPD

R

Kahaietal[22]

Decision

supp

ort

syste

m(D

SS)

143im

ages

Louisia

naState

University

Eye

Center

Grayscale

100

67

mdashNPD

R

Usher

etal[23]

Neuraln

etwork

1273

consecutive

patie

ntsSt

Thom

asHospital

Grayscale

948

mdashmdash

NPD

R

Gardn

eretal[24]

Back

prop

agation

neuralnetwork

200diabetic

and101n

ormal

images

ofprivate

hospita

l

Grayscale

884

835

mdashBo

thNPD

RandPD

R

Reza

andEswaran

[25]

Rulebasedcla

ssifier

STARE

Green

channel

972

100

97

NPD

R

Ann

ieGrace

Vimalaa

ndKa

jamoh

ideen[26]

Cost-e

ffective

compu

ter-aideddiagno

stic

syste

m

Privatee

yeHospital

HSV

916

905

912

NPD

R

Dup

asetal[27]

Automated

fund

usph

otograph

analysis

algorithm

sMessid

orGrayscale

839

727

mdashNPD

R

Ashrafetal[77]

LocalB

inaryPatte

rn

SVM

DIARE

TB1

Green

channel

8748

8599

8615

087

NPD

R

Scientifica 7

Table4Differentm

etho

dsford

etectio

nof

optic

disc

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Accuracy

Rathod

etal[29]

Multilevel2D

wavele

tHistogram

Equalization

HRF

Green

channel

95

Sekara

ndNagarajan

[30]


cluste

ringandhisto

gram

approaches

Messid

orRG

B9958

LuandLim

[31]

Circular

transfo

rmation

lineo

perator

DIARE

TDB0

DIARE

TDB1

DRIVE

STARE

CIEL

AB

974

Acharyae

tal[32]

Otsu

Gradientvectorfl

ow(G

VF)

snakeAtanassov

Intuition

isticFu

zzyHiston

(AIFSH

)Ka

sturbaM

edicalCollege

Grayscale

100

Trucco

etal[33]

Binary

vesselmasks

ared

evelo

pedwith

inVA

MPIRE

softw

are

HRIS(H

igh-Re

solutio

nIm

age

Set)

VDIS

(VascularD

iseaseImage

Set)

Grayscale

957

921

8 Scientifica







Scientifica 9






Microaneurysms

Microaneurysms





10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE


HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Scientifica 7

Table4Differentm

etho

dsford

etectio

nof

optic

disc

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Accuracy

Rathod

etal[29]

Multilevel2D

wavele

tHistogram

Equalization

HRF

Green

channel

95

Sekara

ndNagarajan

[30]


cluste

ringandhisto

gram

approaches

Messid

orRG

B9958

LuandLim

[31]

Circular

transfo

rmation

lineo

perator

DIARE

TDB0

DIARE

TDB1

DRIVE

STARE

CIEL

AB

974

Acharyae

tal[32]

Otsu

Gradientvectorfl

ow(G

VF)

snakeAtanassov

Intuition

isticFu

zzyHiston

(AIFSH

)Ka

sturbaM

edicalCollege

Grayscale

100

Trucco

etal[33]

Binary

vesselmasks

ared

evelo

pedwith

inVA

MPIRE

softw

are

HRIS(H

igh-Re

solutio

nIm

age

Set)

VDIS

(VascularD

iseaseImage

Set)

Grayscale

957

921

8 Scientifica







Scientifica 9






Microaneurysms

Microaneurysms





10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE


HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















8 Scientifica







Scientifica 9






Microaneurysms

Microaneurysms





10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE


HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Scientifica 9






Microaneurysms

Microaneurysms





10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE


HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















10 Scientifica

Table5Differentm

etho

dsford

etectio

nof

exud

ates

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Ram

andSivasw

amy[41]

Clusterin

g-basedmetho

dandcolor

spacefeatures

DIARE

TDB1

RGB

CIE


HSV

HIS

7196

mdash897

Soares

etal[42]

Morph

ologicalop

eratorsa

ndadaptiv

ethresho

lding

DIARE

TDB1

Green

channel

9749

9995

9991

Jayaku

mariand

Santhanam

[43]

Energy

minim

ization

metho

dusingecho

stateneural

network

PrivateH

ospital

mdash90

mdashmdash

Karegowd

aetal[44]

KNNFP

andWKN

NFP

classifiers

DIARE

TDB1

HIS

mdashmdash

9750

WKN

NFP

9667

KNNFP

Ameletal[45]

Com

bine

the119870

-means

cluste

ringalgorithm

and

mathematicalmorph

ology

Oph

thalmologic

Images

CIEL

ab9592

9978

9970

Rokade

andManza

[46]

Haarw

avele

tstransfo

rmation

KNNcla

ssifier

MISP

DIRET

DB0

DIRET

DB1

STARE

Green

channel

3714

218

712502547

mdashmdash

Kayaland

Banerje

e[2]

Medianfilterin

gim

age

thresholding

DIARE

TDB0

DIARE

TDB1

Grayscale

9725

9685

mdash

Jaya

etal[47]

Morph

ologicalop

erations

Circular

Hou

ghtransfo

rm

Fuzzysupp

ortvectorm

achine

PrivateH

ospital

mdash941

900

mdash

Rozlan

etal[48]

Morph

olog

yop

eration

columnw

iseneighb

orho

odso

peratio

nSung

aiBu

loh

Hospital

Green

channel

mdashmdash

60

Soman

andRa

vi[49]

Circular

Hou

ghtransfo

rmandbit

planes

licingmorph

ological

operations

Standard

Diabetic

Retin

opathy

Green

channel

09362

mdash88

Ann

unziatae

tal[50]

Multip

lescaleH

essia

napproach

STARE

HRF

Green

channel

mdashmdash

9562

9581

VanGrin

sven

etal[51]

Bagof

Words

approach

Messid

orEU

GEN

DA

HSV

YCb

Crmdash

mdash090

AUC

Kaur

andMittal[52]

Dyn

amicregion

grow

ing

metho

dSG

HSho

spita

lGrayscale

mdashmdash

9865

Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Scientifica 11

Hemorrhages

Hemorrhages










12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















12 Scientifica

Table6Differentm

etho

dsford

etectio

nof

microaneurysm

s

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Soph

arak

etal[53]


entShadeC

orrection

andextend

edminim

atransform

Patie

ntdata

Green

channel

8161

999

9998

Krish

naetal[54]


sdetectorW

alterK

leinand

CLAC

HE

Messid

orGrayscale

mdashmdash

mdash

Royetal[55]

Cann

yedge

detection

morph

ological

reconstructio

nDIARE

TDB1

Green

channel

895

821

mdash

Adaletal[56]

Con

trastenh

ancementtechn

iqueH

essia

n-based

cand

idates

electio

nalgorithm

and

SVM

classifier

ROC

Green

channel

mdash44

64

mdash

Dattaetal[57]

Con

trastL

imitedAd

aptiv

eHistogram

Equalization

medianfilterandIm

ageC

atenation

Privated

ata

Green

channel

mdash8264

9998

Giancardo

etal[58]

Rado

ncliffop

erator

ROC

Grayscale

41

mdashmdash

DingandMa[

59]

Dyn

amicmultip

aram

eter

template(DMPT

)matching

scheme

ROC

96

mdashmdash

ROC

retin

opathy

onlin

echallenge

Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Scientifica 13

Table7Differentm

etho

dsford

etectio

nof

hemorrhages

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Hatanakae

tal[70]

Gam

mac

orrection

density

analysis

Privated

ata

HSV

80

88

mdash

Tang

etal[71]

Splatfeaturefiltera

pproachand

wrapp

erapproach

Messid

orGrayscale

mdashmdash

096

AUCat

splatlevel

and

087

atim

age

Level

SalehandEswaran

[72]

H-m

axim

atransform

ation

andmultilevel

Thresholding

Privated

ata

Green

channel

8431

and8753

9363

and9508

mdash

Lachuree

tal[73]

SVMK

NNcla

ssifier

Messid

orD

B-rect

HIS

90

100

mdash

Sudh

aand

Thiru

purasund

ari[74]

Medianfilteradaptiv

ehisto

gram

and

KNNClassifi

erMessid

orGrayscale

100

90

mdash

Sharmae

tal[75]

Dyn

amicthresholding

DIARE

TDB1

Green

channel

90

mdashmdash

Akram

etal[76]

GaussianMixture

Mod

elFilter

Bank

DRIVE

STARE

DIARE

TDB

Messid

orGreen

channel

9783

9836

9812

Ashrafetal[77]

LocalB

inaryPatte

rn(LBP

)SV

MDIARE

TDB1

Green

channel

8748

8599

8615

14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















14 Scientifica




8 Future Directions


9 Conclusion


Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Scientifica 15

Table8Differentm

etho

dsford

etectio

nof

bloo

dvessels

Algorith

mIm

agep

rocessingtechniqu

esDatabase

Color

space

Sensitivity

Specificity

Accuracy

Bhatiaetal[28]

GaussianfilterCa

nnyedge

detection

morph

ologicalop

erationsand

Otsu

thresholding

DRIVE

STARE

Green

channel

7031

9735

9523

Fraz

etal[78]

Ensembles

ystem

ofbagged

andbo

osted

decisio

ntre

esG

abor

filterand

morph

ologicaltransfo

rmation

STARE

DRIVE

CHASE

DB1

Green

channel

075

074

072

097

098

097

095

094

094

Nguyenetal[79]

Vesselsegm

entatio

nbasedon

theline

detectorsa

tvarying

scale

STARE

DRIVE

Green

channel

mdashmdash

093 Acc

094 Acc

Fraz

etal[80]

Multiscalelin

edetectio

nmetho

dGabor

filter

DRIVE

STARE

Messid

orGrayscale

073

073

077

097

097

098

094

095

096

Yinetal[81]

Spectralclu

sterin

gtechniqu

ebased

onmorph

ologicalfeaturesH

essia

nmatrix

DRIVE

REVIEW

Green

channel

mdashmdash

Above

94

Vega

etal[82]

Lattice

NeuralN

etworks

with

Dendritic

Processin

gST

ARE

Green

channel

mdashmdash

998

Marın

etal[83]

NeuralN

etwork

DRIVE

STARE

Green

channel

mdashmdash

095

097

Hou

[84]

Multid

irectionalm

orph

ologicaltop-hat

transfo


ment

DRIVE

STARE

Green

channel

073

073

096

096

094

093

Sham

ietal[85]

Morph

ologicalop

erations

Nikoo

kari

Green

channel

8582

9998

mdash

16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















16 Scientifica


Competing Interests


References




























Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Scientifica 17





























18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















18 Scientifica






























Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Scientifica 19

































20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















20 Scientifica











of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of





















of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Documents

Review Article A Review on Recent Developments for ...downloads.hindawi.com/journals/scientifica/2016/6838976.pdf · Images of retina are taken by a device called fundus camera. Retinal