S.W.Franklin,S.E.Rajan/AppliedSoftComputing22(2014)94–100
95
detectionofextravesselgrowth.Forautomaticanalysisofopticfundusthatdetectsthebloodvesselirregularities,segmentationofthevesselsfromthebackgroundismadeinitiallyandtherebyfurtherprocessinglikefeatureextractioncanbeachieved.
Inhumanretina,oneofthemostimportantorgansistheopticnervewhichactsastheconvergentpointofthebloodvesselnet-work.Thecentralretinalarteryandcentralretinalvein owoutthroughtheopticnervethatsuppliesbloodtotheupperlayersoftheretina.Besides,theopticnerveactsasachanneltoconveytheinformationfromtheeyetothebrain.Inearlystages,mostoftheretinalpathologiesaffectslocallyanddoesnotdistresstheentireretina.Butretinalpathologyonorneartheopticnervemayseverelyaffectthevisionevenattheearlystagesbecauseopticnerveisthemostessentialpartforvision[5].Afewobservationsfoundinseveralimportantretinopathiesareattenuationchanges,focalnarrowingandocclusionofretinalarteries.Thediameterandshapeofaretinalvesselplaysakeyroleasindicatorsinophthalmo-logicstudies.Thesechangesgivevaluableinformationtoidentifythesuccessivestagesofdiseasesandtheirresponsetovariousther-apies.Theopticnervecanbeobservedinacloseviewoftheretinalfundus,wheretheopticdiscistheportionofthenervethatisvisi-bleorperceivablebytheeye.Fundusimagingisoneofthepopularclinicalproceduresavailabletorecordthiscloseviewobservationsoftheretina.Thisfundusimagingprocedureisalsousedforthediagnosisandevaluationofthehealthyandnon-healthyretinasofhumaneye[6].Inahealthyretinatheopticnervehasastandardidenti ablesize,shape,colorandlocationrelativetothebloodves-sels.Nevertheless,inaretinacontaininglesions,anyoneormoreofthesepropertiesmaybedeviatedfromitsstandardlevelandshowalargevariance.
Atvariousstagesofthedisease,thevascularnetworkinretinaisverymuchaffectedandhencevariousmorphologicalchangesoccurinretinalvessels.Wecansubstantiallyobserveenoughgeometri-calchangesindiameter,branchingangle,lengthortortuosityintheretinalbloodvesselsduetodiseases.Thesegmentationandmea-surementofretinalbloodvesselscanbeusedtogradetheseverityofcertaindiseases.Thesignofrisklevelfordiabeticretinopathyisthevariationinwidthofretinalvesselswithinthefundus.Oneofthemostimportanttoolsforthepredictionofproliferatedia-beticretinopathyistheabnormalvariationindiameteralongthevein.Moreover,thevariousretinalmicrovascularabnormalitiespredictedareseentobetheearlysymptomsfortheriskofstroke.Inallthesecases,thedesiredfocusisonthevariationindiame-terofthevesselandnotintheexactdiameterofthevessel.Analternativeapplicationofretinalvesselsegmentationisbiometricidenti cationusingdistinctiveretinalvascularnetworkforeachindividual.
Thisresearchpaperisorganizedasfollows:Section“Overviewofretinalvesselsegmentation”reviewsvariousothervesselsegmen-tationmethods,Section“Methodology”describesandillustratesthemethodsofourproposalforvesselsegmentation,Section“Experimentalresultsanalysis”presentsitsresultsandcomparestheperformancewithsomeofthetechniquespreviouslypublishedandSection“Discussionandconclusion”putsforwardthediscus-sionandconcludesthiswork.
Overviewofretinalvesselsegmentation
Segmentationofvesselsinretinalimagesisachievedbyclassify-ingeachimagepixelasvesselornonvesselbasedonthelocalimagefeatures.Intwo-dimensionalimages,thetechniquesadoptedforidentifyingvesselsarebasedonspeci cpropertiesofthevascularsegments.Therearetwobasicapproachesthataregenerallycon-sideredfortheidenti cationofgeneralvascularsegments,whichalsoplayakeyroleinretinalvesselsegmentationapplications.The
algorithmsusedforthesegmentationofbloodvesselsarebroadlyclassi edaspixelprocessing-basedmethodsandvesseltrackingorvectorialtrackingmethods[2].Supplementarymethodsusedforvesselsegmentationinretinalimagescanbeclassi edintotwogroups.The rstgroupconsistsofrule-basedmethodsandthesecondgroupconsistsofsupervisedmethods,wheretrainingofmanuallylabeledimagesisused.Neuralnetworkbasedpixelclas-si cationschemefallsunderthesecondcategory[7,8],whichisfollowedinthispaper.
Pixelprocessing-basedmethodsnormallyconsistoftwophases.Inthe rstphase,anenhancementprocedureisimpliedanditselectsaninitialsetofpixels,whichisfurtherensuredasvesselsinthesecondphase[9].TheretinalvesselsegmentationmethodpresentedinRef.[2]consistsofthreeprocessingphases.Inthe rstphase,backgroundnormalizationofmonochromaticinputimageisperformedandlaterthinvesselenhancementproceduresareused.Inthesecondphase,thevesselcenterlinecandidatepointsareselectedandsubsequentlythesepointsareconnectedandbasedonvessellength,thevalidationofcenterlinesegmentcandidatesisachieved.Inthethirdphase,vesselswithdifferentwidthsareenhancedandprocessedusingbinarymorphologicalreconstruc-tiontechniqueandvessel llingisachievedusingregiongrowingprocess.Soaresetal.haveproposedanalgorithm,whereretinalbloodvesselsaredetectedusingGaborwaveletsbyrepresentingeachpixelbyafeaturevectorandthenbyusingBayesianclassi erwithGaussianmixtures,eachpixelisclassi edaseitheravesselornonvesselpixelandthussegmentationisachieved.Theconceptofmatched lterdetectionwasproposedbyChaudhurietal.[10],wheretwelverotatedversionsof2-DGaussianshapedtemplatesareusedtosearchvesselsegmentsalongallpossibledirections.Theresultantimageproducedisthebinaryrepresentationoftheretinalvasculature.Likewise,segmentationofretinalvesselsarealsoobtainedbymatched lteringapproachesusingglobal[11]orlocalthresholdingstrategies[12].Alsoforthepurposeofvesselbor-dersextraction,differential ltersbasedoneither rst-or-secondorderderivativesareused.AtwostageregiongrowingprocedurewasproposedbyMartinez-Perezetal.wherefeaturesderivedfromimagederivativesareusedinthesegmentationofretinalvessels[13,14].AsstatedinRef.[15],theedgedetection,matched lter-ingandregiongrowingprocedurescanalsobecollectivelyusedfortheautomateddetectionofretinalbloodvessels.Jiangetal.haveproposedatechniqueofadaptivelocalthresholdingbasedontheuseofveri cation-basedmultithresholdschemecombinedwithclassi cationproceduresinRef.[16],usedforthedetectionofves-selsinretinalimages.Atechniqueusedforthesegmentationofvessellikepatternsinretinalimagesthatcombinesmorphological ltersandcross-curvatureevaluationwasproposedbyZanaetal.inRef.[17].Inthisapproach,vesselsegmentsaloneareconsideredasimagefeatureandhenceusingmorphological lterssimpli esthecomputationofcross-curvature.Analgorithmforretinalves-selsegmentationbasedontheclassi cationofpixelsusingsimplefeaturevectorwasproposedbyNiemeijeretal.inRef.[18].Anewmethodofsegmentationofbloodvesselsinretinalimageswaspro-posedbyStaaletal.inRef.[19].Thissupervisedmethodiscalledprimitive-basedmethodandthisalgorithmisbasedontheextrac-tionofimageridgesusedtocomposeprimitivesthatdescribethelinearsegmentscalledlineelements.Thepixelsthoseareassignedtotheclosestlineelementpartitionstheimageintheformofimagepatchesandareclassi edusingasetoffeaturesfromthecorre-spondinglineandimagepatch.Inaddition,atechniquebasedonneuralnetworkisusedtoidentifytheretinalbloodvessels,wheretheinputsareobtainedusingprincipalcomponentanalysisandthenedgedetectiontechniqueisused.
Vesseltrackingmethodsusetheconceptofmeasuringsomelocalimagepropertiestolocatethevesselpointswhichareusedfortracingthevasculature.Inthesetypesofalgorithms,boththe
ICPR06_A new approach to automated retinal vessel segmentation using ...Gabor filters and scale production, and a threshold probing technique ...
热门文档