彩色视网膜图像计算机辅助检测DR解剖结构和病变的研究进展(IJIGSP-V7-N11-8)

I.J.Image,GraphicsandSignalProcessing,2015,11,55-69PublishedOnlineOctober2015inMECS()DOI:10.5815/ijigsp.2015.11.08Copyright©2015MECSI.J.Image,GraphicsandSignalProcessing,2015,11,55-69AReviewofComputerAidedDetectionofAnatomicalStructuresandLesionsofDRfromColorRetinaImagesSreejiniKSandV.KGovindanDepartmentofComputerScienceandEngineering,NITCalicut,Kerala,IndiaEmail:[e-mail],[e-mail]—Ophthalmologyisthestudyofstructures,functions,treatmentanddisordersofeye.Computeraidedanalysisofretinaimagesisstillanopenresearcharea.Numerouseffortshavebeenmadetoautomatetheanalysisofretinaimages.Thispaperpresentsareviewofvariousexistingresearchindetectionofanatomicalstructuresinretinaandlesionsforthediagnosisofdiabeticretinopathy(DR).Theresearchindetectionofanatomicalstructuresisfurtherdividedintosubcategories,namely,vesselsegmentationandvesselcenterlineextraction,opticdiscsegmentationandlocalization,andfovea/maculadetectionandextraction.Variousresearchworksineachofthecategoriesarereviewedhighlightingthetechniquesemployedandcomparingtheperformancefiguresobtained.Theissues/lacunaofvariousapproachesarebroughtout.Thefollowingmajorobservationsaremade:Mostofthevesseldetectionalgorithmsfailtoextractsmallthinvesselshavinglowcontrast.Itisdifficulttodetectvesselsatregionswhereclosevesselsaremerged,atregionsofmissingofsmallvessels,atopticdiscregions,andatregionsofpathology.Machinelearningbasedapproachesforbloodvesseltracingrequireslongprocessingtime.Itisdifficulttodetectopticdiscradiusorboundarywithsimplebloodvesseltracing.Automaticdetectionoffoveaandmacularregionextractionbecomescomplicatedduetonon-uniformilluminationswhileimaginganddiseasesoftheeyes.Techniquesrequiringpriorknowledgeleadstocomplexity.Mostlesiondetectionalgorithmsunderperformduetowidevariationsinthecoloroffundusimagesarisingoutofvariationsinthedegreeofpigmentationandpresenceofchoroid.IndexTerms—Diabeticretinopathy,retinafundusimage,opticdisc,macula,fovea,bloodvessels,exudates,microaneurysms,hemorrhages,segmentation,survey.I.INTRODUCTIONAbnormalitiesinretinaaretheindicatorsofvariousdiseasesinhumanbodyandresearchinthisfieldbecamepopularduetotheincreaseofprevalenceofdiabeticpatients.Diabetesisalifestylediseasewhichinterfereswiththeabilityofhumanbodytostoreandproduceinsulin.Thispancreasgeneratedinsulincontrolsthesugarlevelofblood[1].IthasadverseeffectontheeyescalledDiabeticRetinopathy(DR),onnervoussystemcalleddiabeticneuropathy,onkidneyscalleddiabeticnephropathy,butmostlikelyaffectedpartisretinaisthepatientvision.TheaftereffectofDRisthedamageofretinalbloodvessels.DRhasnoearlysigns,thusmostofthepatientsareunawareoftheirdiseaseuntilitturnstobecomemoresevere[2].However,itsadverseeffectonvisionmaybeavoidedbyconsultingophthalmologistforregularscreeningforDR.Ophthalmologistshavetoputupgoodefforttoanalyzefundusimagesandarriveatavalidconclusionregardingtheseverityofthedisease.Iftheinitialtaskssuchasanalyzingtheimagesandidentificationofvariouslesionsandvesselsegmentationareautomated,theDoctorscanmakeaquickdiagnosis.Thus,aninexpensiveautomatedsystemcapableofmeetingthedemandsofincreasingdiabeticpopulationisofgreathelpforDoctorsanddiabeticpatients.Fig.1.Lossofvisionduetotheinfluenceofdiabetes:(a)Normalvision(b)DiabeticRetinopathy(Courtesy:NationalEyeInstitute,NationalInstituteofHealth[3]).TheimagesshowninFig.1demonstratetheadverseeffectofDRonpatientvision.ThesampleretinalimagewithmarkedlesionsandmainretinalfeaturesisgiveninFig.2.TheimageofretinawastakenfromDIARETDB1(a)Healthyretina(b)AbnormalretinaFig.2.Retinafundusimages(a)Normalhealthyretina(b)DiabeticRetinopathy[4].56AReviewofComputerAidedDetectionofAnatomicalStructuresandLesionsofDRfromColorRetinaImagesCopyright©2015MECSI.J.Image,GraphicsandSignalProcessing,2015,11,55-69dataset[3].Visionlosswilloccurwhenexudatesarepresentinmacularregion,whichisknownasDiabeticMacularEdema(DME).Fig.2.ashowsthenormalhealthycolorretinaimage.AbnormalretinaimagewherevariouslesionsandmainretinalstructuresmarkedispresentedinFig.2.b.Beforewereviewtheapproaches/techniquesfordetectionofanatomicalstructuresofretinaandlesionsofDR,wepresentbriefintroductionstoretinalfundusimaging,anatomicalstructuresofretina,lesionsofdiabeticretinopathyandpublicallyavailableretinaldatabasesinthefollowingsubsections:A.RetinalFundusImagingRetinalimagesareacquiredthroughvarietyoftechniquessuchasFluoresceinAngiograms(FA)orfunduscamera.Fluoresceinisinjectedintoourbodybeforeimagingforimprovingthecontrastoffeatureslikearteries,capillariesandveins.Imagesthusobtainedwillhavebettervisibilityforvesselfeatures.However,suchgray-scaleimagingofretinaistimeconsuming,inconvenient,costlyandhavingsideeffectsonpatients.Asthegrayscaleimagesdonotprovidealldetailsofretina,itisnotadequateforproperdiagnosis.Nowadays,useofcolorfundusimagestakenthroughfunduscameraisfoundusefulforbetterdiagnosis.ThismakesuseofcolorCCDsensorstograbfinedetailsofretinaimage.B.AnatomicalStructuresofRetinaOpticdiscorblindspot–opticnervesthatenterintotheroundzoneregionoftheret