Classification of NPDR.ppt

Detection and Classification of Non-Proliferative Diabetic Retinopathy Stages Using Morphological Operations and SVM Classifier

1. Detection and Classification of Non-
Proliferative Diabetic Retinopathy Stages
Using Morphological Operations and SVM
Classifier
Your logo here
Mohammed Shafeeq Ahmed
Research Scholar,
Dept. of Computer Science,
Research and Development Center, Bharathiar University, Coimbatore

2. Your logo here
Overview
1.Introduction
1.What is Diabetic Retinopathy
2.Classification of Diabetic Retinopathy
2.Material Collected
3.Methodology
4.Experimental Results
5.Conclusion
6.References

3. •Peopleswithdiabetescanhaveaneyediseasecalleddiabeticretinopathy.
•Thisiswhenhighbloodsugarlevelscausedamagetobloodvesselsinthe
retina.Thesebloodvesselscanswellandleak.Ortheycanclose,stopping
bloodfrompassingthrough.Sometimesabnormalnewbloodvesselsgrow
ontheretina.Allofthesechangescanstealyourvision.
•Diabeticretinopathyisatermusedforalltheabnormalitiesofthesmall
bloodvesselsoftheretinacausedbydiabetes,suchasweakeningofblood
vesselwallsorleakagefrombloodvessels.
•Diabeticretinopathyisacomplicationofdiabetesthatresultsfromdamage
tothebloodvesselsofthelight-sensitivetissueatthebackoftheretina.
•Diabeticretinopathyaffectanyonewhohastype1diabetesortype2
diabetes.Thelongerapatienthasdiabetes,themorelikelytheyareto
developdiabeticretinopathy.
Your logo here
INTRODUCTION
What is Diabetic Retinopathy?

4. •Therearetwomainstagesofdiabeticeyedisease.
•Non-proliferativeretinopathyisacommon,usuallymildformthatgenerally
doesnotinterferewithvision.Abnormalitiesarelimitedtotheretinaand
usuallywillonlyinterferewithvisionifitinvolvesthemacula.Ifleftuntreated
itcanprogresstoproliferativeretinopathy
•Proliferativeretinopathy,themoreseriousform,occurswhennewblood
vesselsbranchoutorproliferateinandaroundtheretina.Itcancausebleeding
intothefluid-filledcenteroftheeyeorswellingoftheretina,andleadto
blindness.
Your logo here
Diabetic Retinopathy
A normal retina
A retina showing a sign
of diabetic retinopathy

5. Thisistheearlystageofdiabeticeyedisease.Manypeoplewithdiabeteshave
it.
Stage1:MildNon-proliferativeRetinopathy.
Atthisearlieststage,microaneurysmsoccur.Theyaresmallareasofballoon-
likeswellingintheretina'stinybloodvessels.
Stage2:ModerateNon-proliferativeRetinopathy.
Asthediseaseprogresses,somebloodvesselsthatnourishtheretinaare
blocked.
Stage3:SevereNon-proliferativeRetinopathy
Manymorebloodvesselsareblocked,deprivingseveralareasoftheretina
withtheirbloodsupply.Theseareasoftheretinasendsignalstothebodyto
grownewbloodvesselsfornourishment.
Stage4:ProliferativeRetinopathy.
Atthisadvancedstage,thesignalssentbytheretinafornourishmenttrigger
thegrowthofnewbloodvesselswhichareabnormalandfragile.Theygrow
alongtheretinaandalongthesurfaceoftheclear,vitreousgelthatfillsthe
insideoftheeye.Bythemselves,thesebloodvesselsdonotcausesymptoms
orvisionloss.However,theyhavethin,fragilewalls.Iftheyleakblood,
severevisionlossandevenblindnesscanresult.
Your logo here
Classification of Non-Proliferative
Diabetic Retinopathy (NPDR)

6. Theimplementationofthedetectionmethodproposedwasperformedin
MATLAB.Theaccuracyofthemethodwastestedinpublicdatabaseoffundus
imagesDIARETDB1.TheDIARETDB1hasatotalof89(RGB)fundus
imagesofsize1500x1152.Outofthistotal,84imageshavecharacteristic
signsofDR,suchasmicroaneurysms,hemorrhagesandexudates,and5
imagesareofnormalretinas.
Furtherwehaveconductedtheexperimentondifferentdimensionlike(4288x
2848),(786x584),capturedfromdifferentcameraandDRIVEdataset.To
checktheperformanceoftheproposedalgorithm.
Your logo here
Material Collection

7. Your logo here
Methodology
Feature Extraction
SVM Classifier
Pre-processing
Applying Morphological Method
Input Fundus Images
Mild Moderate Severe
Work Flow of NPDR Classification
•Weproposedanefficientmethodfor
automaticdetectionandclassificationof
NPDR stagesusingmorphological
operationsandSVMclassifier.
•Weperformexperimentsonalarge
datasetcollectedfromthepublicly
availableDIARETDB1andDRIVE
database.
•Intheproposedmethod,bloodvessels,
MAsandexudatesareusedtodetectDR
stages.Thetechniqueusedisbydividing
thefundusimageintofourquadrantsand
theareaofMAsandexudatesiscomputed
ateachofthefourquadrants.SVM
classifierisusedtoidentifythedifferent
stagesofNPDRinfundusimage.

8. Preprocessing:
•Inputimagesisconvertedtoastandardsizetoimprovethequalityofinput
image.
•GreenchannelisusedforfeatureextractionofDRfromfundusimages.
•ContrastenhancementoftheoutputimageisdoneusingCLAHEapproach.
BloodVesselDetection:
•Morphologicaloperationisusedforextractionofbloodvesselsanddetection
andeliminationofOpticDisc.
•SegmentationisusedtoeliminateotherfeatureslikeMAsandexudates.
•Lasttheareaofbloodvesselsiscalculated.
•Bloodvesselsareaismoreinaffectedretinaascomparetothenormalretina.
•Henceweclassifiesasmild,moderateandsevereNPDR.
MicroaneurysmsDetection:
•EdgedetectionisachievedbyusingCannyEdgedetector.
•Thresholdingtechniqueisusedtoeliminatenoiseandexudates.
•Theresultantimageisdividedintofourquadrantsandcomputethearea
occupiedbyMAsineachquadrants.
Your logo here

9. ExudatesDetection:
•ExudatesaredetectedusingMorphologicaloperations.
•Normalretinalimagedoesnotcontainanyexudates.
•MildNPDRmaycontainexudates,butonlyinonequadrant.
•ModerateNPDRareaffectedwithexudatesanddistributedinatleasttwo
quadrants.
•SevereNPDRareaffectedwithnumerousexudatesandpresentinalmostall
thequadrants.
Your logo here

10. •Theproposedmethodhasbeenevaluatedusing129fundusimages
collectedfromtheDIARETDB1andDRIVEdatabase.
•AllthethreefeaturesofDRhavebeendetectedsuccessfully.
•InthenormalimagesthebloodvesselsoccupythelargerareaandMAsand
exudatesareabsent.
•InthecaseofmildNPDRandmoderateNPDR,theMAsandexudates
showedtheirpresenceandinsevereNPDRtheirprominenceismore.
•TheSVMclassifierhasbeenusedforclassification.
•Thefeatureswereclassifiedasnormal,mildNPDR,moderateNPDR,and
severeNPDR.
•Anaverageaccuracyof100%,93.33%,100%and86.67%isobtainedfor
normal,mildNPDR,moderateNPDR,andsevereNPDR,respectively.The
sensitivityof96.08%andspecificityof97.92%isobserved.Thedetailsof
theclassificationobtainedarepresentedinTable1.
•Drawback:TherecognitionresultsinthecaseofseverNPDRislow
comparedtootherstagessincemanyofthefundusimageswithsevere
NPDRweremisclassifiedasmoderateNPDR.
Your logo here
Experimental Results

11. Your logo here
No MAs or Exudates Detected
(NORMAL) Fundus Images

12. Your logo here
Mild NPDR Detected Fundus Images

13. Your logo here
Moderate NPDR Detected Fundus
Images

14. Your logo here
Sever NPDR Detected Fundus
Images

15. •AnautomatedsystemwasdevelopedtodetectandclassifytheNPDR
stagesfromthefundusimagesusingSVMclassifierandachievedahigh
percentageofsensitivityandspecificity.
•Theadoptedmethodologyismoreefficientandeffectivethanother
methods.
•Wehaveused129fundusimagescollectedfromtheDIARETDB1and
DRIVEdatabases.
•TheperformanceofthemethodfordetectionandclassificationofNPDR
stagesfromfundusimageshasachievedahighsuccessfulpercentage.
•Thismethodishighlyflexibletootherdatabases.
•ThesystemhasclassifiedtheNPDRstagesinnormal,mildNPDR,
moderateNPDRandsevereNPDRwithanaverageaccuracyof95%,an
averagesensitivityof96.08%andanaveragespecificityof97.92%.
•Themainpurposeofthisproposedworkistohelptheophthalmologistsin
screeningtheDRusingtheSVMclassifier.
•Thus,thisSVMtechniquehasgivenasuccessfulDRscreeningmethod
whichhelpstodetectthediseaseinearlystageandpreventsvisionloss.
Your logo here
Conclusions

16. 1.ReportontheExpertCommitteeontheDiagnosisandClassificationofDiabetesMellitus,DiabetesCare,July1997.
2.AmericanDiabetesAssociation,DiagnosisandClassificationofDiabetesMellitusDiabetesCare,Jan2013.
3.SicreeR,ShawJ,ZimmetP.Diabetes&ImpairedGlucoseTolerance.BakerIDIHeartandDiabetesIns.IDFDiabetesAtlas.4
th
ed.
DiabetesandIGT,2009.
4.RamachandranA,SnehalathaC,KapurA,VijayV,MohanV,DasAK,etal.DiabetesEpidemiologyStudyGroupInIndia(DESI):High
prevalenceofdiabetesandimpairedglucosetoleranceinIndia:NationalUrbanDiabetesSurvey.Diabetologia2001.
5.MohanV,MathurP,DeepaR,DeepaM,ShuklaDK,MenonGR,etal.Urbanruraldifferencesinprevalenceofself-reporteddiabetesin
India:TheWHO-ICMRIndianNCDriskfactorsurveillance.DiabResClinPract2008.
6.ChowCK,RajuPK,RajuR,ReddyKS,CardonaM,CelermajerDS,etal.TheprevalenceandmanagementofdiabetesinruralIDC
2006.
7.http://www.icoph.org/downloads/Diabetic-Retinopathy-Detail.pdf.
8.DWelfer,J.Scharcanski,D.R.Marinho,Acoarse-to-finestrategyforautomaticallydetectingexudatesincoloreyefundusimages,
ComputerizedMedicalImagingandGraphics,2010.
10.URAcharya,CMLim,EYKNg,CCheeandTTamura,Computer-baseddetectionofdiabeticretinopathystagesusingdigitalfundus
images:Parth:journalofengineeringinMedicine,2009.
11.Niemeijer,M,vanGinneken,B.,Staal,J.,Suttorp-Schulten,M.,andAbramoff,M.Automaticdetectionofredlesionsindigitalcolor
fundusphotographs.IEEEtransmedimaging,2005.
12.JNayak,PSBhat,URAcharya,CMLim,andMKagathi,AutomatedIdentificationofDiabeticRetinopathyStagesUsingDigital
FundusImages.J.Med.Systems,2008.
13.H.LiandO.Chutatape,“Automatedfeatureextractionincolorretinalimagesbyamodelbasedapproach,”IEEETrans.onMedical
Engineering,2004.
17.C.I.Sanchez,M.Garcia,A.Mayo,M.LopezandR.Hornero,“Retinalimageanalysisbasedonmixturemodelstodetecthardexudates,”
MedicalImageAnalysis,2009.
18.AkaraSopharak,BunyaritUyyanonvara,SarahBarmanproposedan“Automaticexudatesdetectionfordiabeticretinopathyscreening”,
2009.
19.http://www.it.lut.fi/project/imageret/diaretdb1/
20.http://www.isi.uu.nl/Research/Databases/DRIVE
21.MohammedShafeeqAhmedandDr.B.Indira,“Automaticdetectionandidentificationofexudatesinretinalfundusimagesusing
morphologicaltechniques”,InternationalConferenceonRecentTrendsinEngineeringandMaterialSciences,2016.
22.RafaelC.Gonzalez,RichardEugeneWoods,StevenL.Eddins“DigitalimageprocessingusingMatlab”PearsonEducationIndia.
Your logo here
Reference

17. Your logo here
Thank You

18. Your logo here