FluidsegmentationinNeutrosophicdomain
ElyasRashno
DepartmentofComputerEngineering IranUniversityOfScienceandTechnology Tehran,Iran elyas.rashno@gmail.com
AbdolrezaRashno
DepartmentofComputerEngineering EngineeringFaculty LorestanUniversity Khorramabad,Iran rashno.a@lu.ac.ir
SadeghFadaei DepartmentofElectricalEngineering FacultyofEngineering YasoujUniversity Yasouj,Iran,7591874934 s.fadaei@yu.ac.ir
Abstract—Opticalcoherencetomography(OCT)asretina imagingtechnologyiscurrentlyusedbyophthalmologistas anon-invasiveandnon-contactmethodfordiagnosisofagerelateddegeneration(AMD)anddiabeticmacularedema(DME) diseases.FluidregionsinOCTimagesrevealthemainsignsof AMDandDME.Inthispaper,anefficientandfastclustering inneutrosophic(NS)domainreferredasneutrosophicC-means isadaptedforfluidsegmentation.Forthistask,aNCMcost functioninNSdomainisadaptedforfluidsegmentationandthen optimizedbygradientdescendmethodswhichleadstobinary segmentationofOCTBscanstofluidandtissueregions.The proposedmethodisevaluatedinOCTdatasetsofsubjectswith DMEabnormalities.Resultsshowedthattheproposedmethod outperformsexistingfluidsegmentationmethodsby6%indice coefficientandsensitivitycriteria.
IndexTerms—Opticalcoherencetomography;Neutrosophic theory;Fluidsegmentation;Retinadiseases
I.INTRODUCTION
Ophthalmologistsuseopticalcoherencetomography(OCT) asanon-contactmethodtodiagnosisandfollowupof retinadisease.FluidregionsaredetectableinOCTimages andrevealabnormalitiesofretinadiseaseincludingdiabetic maculaedema(DME)andage-relatedmaculardegeneration (AMD)[1].Imageprocessinganddataminingapplications havebeenwidelyusedinmanyapplication[2]–[7].Fluid regionscanbesegmentedautomaticallybyimageprocessing anddataminingtools.Manyfluiddetectionmethodshave beenproposedintheliteraturesuchaskernelregression basedsegmentationofOCTimageswithDME[8],fluid-filled regiondelineationofboundariesinOCT[9],semi-automatic segmentationmethodforretinalOCTimagestestedinpatients withdiabeticmacularedema[10],labelpropagationand higher-orderconstraint-basedsegmentationoffluidregionsin OCTimagesofDMEsubjects[11],computerizedassessment ofintraretinalandsubretinalfluidregionsinOCTimages oftheretina[12],automatedsegmentationofpathological cavitiesinOCTscans[13],three-dimensionalanalysisof retinallayerandfluid-filledregionsinOCTofthemacula [14],probabilityconstrainedgraph-search-graph-cutmethod forthree-dimensionalsegmentationoffluid-associatedabnormalities[15],fluidsegmentationwithshortest-pathmethods andneutrosophic(NS)theory[16]–[21],fluidsegmentation withdeepconvolutionalnetworks(CNNs)[22]–[25].
SmarandacheproposedNStheoryin1995,abranchofphilosophywhichstudiestheorigin,natureandscopeofneutralities,aswellastheirinteractionswithdifferentideationalspectra[26].Thistheoryhasbeenappliedonmanyapplications includingimagesegmentation[27]–[30],imagethresholding [31],imageedgedetection[32],content-basedimageretrieval [33]–[37],retinalimageanalysis[16]–[20],speechprocessing [38],dataclustering[39],[40]anduncertaintyhandling[41], [42].Themaincontributionofthisworkistooptimizeacost functioninNSdomaintobinarizeretinaOCTscanstofluid andtissueregions.Thecostfunctionisderivedfromfuzzycmeans(FCM)andisoptimizedbygradientdescendmethods insubsequenceiterations.Therestofthispaperisorganized asfollows:SectionIIreviewsFCMandNStheory.Section IIIpresentstheproposedmethod.Experimentalresultsand conclusionarediscussedinsectionsIVandV,respectively.
II.REVIEWOF NS AND FCM
SincethecostfunctioninthisworkisanextensionofFCM andisdefinedinNSdomain,here,areviewonFCMandNS isdiscussedshortly.
A.FCM
TheFCMisaclusteringmethodthatassignsamembership degreeininterval[0-1]toeachdatapoint,Therefore,each datapointisassignedtoallclusterswithdifferentmembership degreesandthesumofmembershipstoallclustersshouldbe 1.FCMcostfunctionisdefinedasfollows:
tothe
th clusterwithcenter v
.
and m areanorm metricandaconstantvariablefordeterminingthefuzziness oftheresultingpartition,respectively.MinimizingFCMcost functionleadstothefollowingequationsforthecomputation ofmembershipdegreesandclustercenters:
Firstly,membershipdegreesareinitializedrandomlyand clustercentersarecomputedbyEq.(3).Then,membership degreesandclustercentersarecomputedrepeatedly,untilthere isnosignificantchangesoftheseparametersinsubsequent iterations.
B.Neutrosophic
NStheoryisanewlybroughtupbranchesofphilosophy.In thistheory,setA,Anti-A(contemplatesAinacorrelationwith itsopposite),Neut-A(neutralityofA,neitherAnorAnti-A) aredefined[26].Ineachapplication,NSsetsshouldbedefined foralldatapointsaswellasrelationsbetweensets.Forexample,inourproblemofinterestasimagesegmentation,each pixelinNSdomainismodeledas P (t,i,f ) meaningthatitis t percenttrue, i percentindeterministic,and f percentfalse. Therefore,threesets PNS (i,j)= {T (i,j),I(i,j),F (i,j)} are definedforpixelsinNSdomainasfollows[27]:
T (i,j)= g(i,j) gmin gmax gmin (4)
g(i,j)= 1 w × w
i+ w 2 m=i w 2
j+ w 2 n=j w 2
g(m,n) (5)
I(i,j)= δ(i,j) δmin δmax δmin (6)
δ(i,j)= |g(i,j) g(i,j)| (7)
F (i,j)=1 T (i,j) (8)
where, g(i,j) representstheintensityvalueofthepixel P (i,j), g isimagematrix g filteredbyaveragefilterwith windowsize w.Differenceoftwomatrixes g and g is computedandstoredin δ
III.PROPOSED METHOD
Inthiswork,NCMcostfunctionisusedandadaptedfor OCTfluidsegmentation.SinceOCTscanscontainnoiseand thereisambiguitybetweenfluidpixelsandmanypixelsin backgroundandtissue,indeterminacyset I inNSdomaincan beveryusefulandmodelssuchambiguitieseasily.Here,each pixelinOCTscansaretransferredtothreesets T , I and F by Eqs.(4),(6)and(8),respectively.Then,NSsetsarepresented forclusteringcostfunctionasfollows:
(T,I,F,C
Cqi = argmax(Tij ) ,j = pi ∩ j =1, 2, ,C (12)
Gradientdescentmethodisusedforcostfunctionoptimizationwhichleadstothefollowingrelationsforthecomputation of T , I and F setsinNSdomainandclustercenters.
Tij = K w1 (Xi Cj ) ( 2 m 1) (13)
Ii = K w2 (Xi Cimax ) ( 2 m 1) (14)
Fi = K w3 δ ( 2 m 1) (15) Cj = N i=1(¯ w1Tij )mXi N i=1(¯ w1Tij )m (16)
K = 1 w1
C j=1 (Xi Cj ) ( 2 m 1) + 1 w2 (Xi Cimax ) ( 2 m 1) + 1 w3 δ ( 2 m 1) 1 (17)
Parameter K isacommonpartinEqs.(13)-(15).Itis computedonceandusedthreetimeswhichleadstospeedupin eachiterationofoptimization.Inthefirststep, 12 clustersare consideredforclustering.Thereasonisthatthereis 11 layers inretinaandeachlayercanbeassignedtooneclusterbasedon itstextureandgraylevel.Oneextraclusterisconsideredfor fluidregionsasfluidcluster.Afterconvergence,clustersare sortedascendinglybasedonthegraylevelofclustercenters. Thefirstclusterisconsideredasfluidclusterwithlabel1and otherclustersareconsideredastissueclusterswithlabel 0 Therefore,theproposedclusteringschemeleadstoabinary segmentationofOCTscans.
IV.EXPERIMENTAL RESULTS
A.EvaluationMetrics
Toshowtheeffectivenessoftheproposedmethod,ithas beenevaluatedwiththreemetrics;dicecoefficient,precision andsensitivity;computedfromtruepositive(TP)fluidpixels detectedasfluid,falsepositive(FP)tissuepixelsdetectedas fluid,truenegative(TN)tissuepixelsdetectedastissueand falsenegative(FN)fluidpixelsdetectedastissue.
Dice Coeff = 2TP 2TP + FP + FN (18)
Sensitivity = TP TP + FN (19)
Precision = TP TP + FP (20)
Thesecriteriaareusedtoevaluatefluidsegmentationresultsofautomatedmethodsincomparisonwithmanually segmentedfluidregionsbyophthalmologistexperts.
B.Dataset
Theproposedfluidsegmentationmethodistestedona datasetfromOPTIMACystSegmentationChallengewhich contains 4 subjectswith 49 imagespersubjectwheretheimage resolutionvariesfrom 512 × 496 to 512 × 1024.Thefluid regionsofeachOCTimagearemanuallysegmentedbytwo ophthalmologistexpertsasgroundtruthimages.Thisdataset ispubliclyavailableandcanbefoundonline1
C.Resultes
Segmentedfluidregionsbytheproposedmethodarecomparedwithgroundtruthimages.Fig.1depictsfluidregions segmentedbyexpertandtheproposedmethodfortwoscans, eachscaninonerow.Theproposedfluidsegmentationmethod iscomparedwith4fluidsegmentationmethodsproposedin [17]and[44]–[46].Theresultsofthesemethodsareshownin Fig.2.Itisclearvisuallythattheproposedmethodhaslower numberoffalsenegaiveswhichleadstohigheraccuracy.
Fig.2:Segmentedfluidby:(a)expert,(b)methodin[17], (c)methodin[44],(d)methodin[45]and(e)theproposed method
TABLEI:Dicecoefficient,sensitivityandprecisionofthe proposedmethodandothermethodsincomparisonwithexpert 1.
Expert1 Sub. Method in[44] Method in[45] Method in[46] Method in[17] Prop. Method
DiceCoeff
Sensitivity
1 73.49 80.43 71.40 82.96 83.17 2 73.90 55.10 45.49 78.11 74.61 3 78.46 75.35 69.54 82.23 84.85 4 78.12 71.78 71.15 80.75 86.32 Ave. 75.99 70.66 64.39 81.01 82.23
1 70.81 82.19 72.49 84.43 86.95 2 96.79 99.04 70.45 98.94 98.42 3 75.72 85.13 47.38 85.18 86.58 4 78.78 80.59 77.79 84.49 90.51 Ave. 80.52 86.73 67.02 88.26 90.59
1 93.00 85.06 54.87 84.03 82.12
Fig.1:Fluidsegmentationresultsfor2scans:(a)InputOCT image,(b)manuallysegmentatedbyexpertand(c) segmentedbytheproposedmethod.



Precision
2 74.36 54.18 51.12 78.48 69.31 3 94.89 79.88 30.93 85.45 83.43 4 96.97 88.62 54.98 93.20 75.01 Ave. 89.80 76.93 47.97 85.29 77.46
TableIreportsthedicecoefficient,sensitivityandprecision oftheproposedmethodandmethodsin[17]and[44]–[46]. ReportedresultsinTableIarethecomparisonbetweenall methodsandfluidregionsmanuallysegmentedbyexpert1. TableIIreportsthesameresultsbutincomparisonwithexpert 2.Incomparisonwiththeresultsofexpert1,theproposed methodoutperformsothermethodswithrespecttodicecoefficientandsensitivityof 82.23% and 90.59%,respectively. Thesemeasuresoftheproposedmethodincomparisonwith expert2are 81 62% and 90 11% whichalsooutperforms othermethods.Forprecisionmeasure,methodsin[17]with 85 29% and[44]with 90 87% havethebestperformancein comparisonwithexpert1and2,respectively.
(a) (b) (c) (a) (b) (c)
V.CONCLUSION
(a) (b) (c) (d) (e)
Thispaperpresentedafluidsegmentationmethodbased onNCMcostfunctioninNSdomain.Minimizingthecost
1https://optima.meduniwien.ac.at/research/challenges/
functionresultedinbinarysegmentationofOctimagesinto tissueandfluidregions.SegmentationresultsonOptima datasetshowedthattheproposedmethodoutperformsother segmentationmethodsindicecoefficientandsensitivitymeasureswhileforprecisioncriteria,othermethodshadthebest performance.Futureeffortswillbedirectedtowardsproposing acostfunctioninNSdomainforbettermodelingofnoiseand uncertaintyinOCTpixels.NScanmodeluncertaintyofOCT pixelsindeepconvolutionalnetworksandleadstomorerobust networkagainstnoiseandimagingdevice.Therefore,using NStheoryforfluidsegmentationbyCNNcanbeproposedas anotherfuturework.
REFERENCES
[1] Kafieh,R.,Rabbani,H.,Abramoff,M.D.andSonka,M.,”Intraretinal layersegmentationof3Dopticalcoherencetomographyusingcoarse graineddiffusionmap,” Medicalimageanalysis,vol.17,no.8,pp.907928,2013.
[2] A.Rashno,B.Nazari,S.Sadri,M.Saraee,Effectivepixelclassification ofmarsimagesbasedonantcolonyoptimizationfeatureselectionand extremelearningmachine,Neurocomputing226(2017)66–79.
TABLEII:Dicecoefficient,sensitivityandprecisionofthe proposedmethodandothermethodsincomparisonwithexpert 1.
Expert2 Sub. Method in[44] Method in[45] Method in[46] Method in[17] Prop. Method
DiceCoeff
1 72.96 79.10 68.17 82.90 83.03
2 71.68 55.11 45.81 79.09 78.12
3 82.33 79.34 65.01 80.36 82.22
4 77.91 71.56 72.55 80.87 83.12 Ave. 76.22 71.27 62.88 80.80 81.62
1 69.95 78.56 66.75 80.94 83.35
2 92.25 94.54 64.71 94.45 93.86
Sensitivity
Precision
3 81.49 90.95 54.84 90.75 92.15
4 78.54 80.22 77.56 83.70 91.06 Ave. 80.55 86.06 65.96 87.46 90.10
1 95.71 86.55 59.61 87.53 82.62
2 74.45 54.14 51.34 78.58 67.79
3 96.10 79.96 37.99 85.48 83.14
4 97.24 88.99 59.50 93.58 77.98 Ave. 90.87 77.41 52.11 86.29 77.88
[3] A.Rashno,H.SadeghianNejad,A.Heshmati,Highlyefficientdimension reductionfortext-independentspeakerverificationbasedonrelieff algorithmandsupportvectormachines,InternationalJournalofSignal Processing,ImageProcessingandPatternRecognition6(1)(2013)91–108.
[4] A.Rashno,S.M.Ahadi,M.Kelarestaghi,Text-independentspeaker verificationwithantcolonyoptimizationfeatureselectionandsupport vectormachine,in:20152ndInternationalConferenceonPattern RecognitionandImageAnalysis(IPRIA),IEEE,2015,pp.1–5.
[5] A.Rashno,M.Saraee,S.Sadri,Marsimagesegmentationwithmost relevantfeaturesamongwaveletandcolorfeatures,in:2015AI& Robotics(IRANOPEN),IEEE,2015,pp.1–7.
[6] A.Rashno,F.S.Tabataba,S.Sadri,Imagerestorationwithregularization convexoptimizationapproach,JournalofElectricalSystemsandSignals 2(2)(2014)32–36.
[7] A.Rashno,F.S.Tabataba,S.Sadri,Regularizationconvexoptimization methodwithl-curveestimationinimagerestoration,in:20144th InternationalConferenceonComputerandKnowledgeEngineering (ICCKE),IEEE,2014,pp.221–226.
[8] Chiu,S.J.,Allingham,M.J.,Mettu,P.S.,Cousins,S.W.,Izatt,J.A.and Farsiu,S.,”Kernelregressionbasedsegmentationofopticalcoherence tomographyimageswithdiabeticmacularedema,” Biomedicaloptics express,vol.6,no.4,pp.1172-1194,2015.
[9] FernandezDC,”Delineatingfluid-filledregionboundariesinoptical coherencetomographyimagesoftheretina,” IEEEtransactionson medicalimaging,vol.24,no.8,pp.929945,2005.
[10] HuangY,DanisRP,PakJW,LuoS,WhiteJ,ZhangX,etal., ”Developmentofasemi-automaticsegmentationmethodforretinaloct imagestestedinpatientswithdiabeticmacularedema,” PloSone,vol. 8,no.12,p.e82922,2013.
[11] WangT,JiZ,SunQ,ChenQ,YuS,FanW,etal.,”Labelpropagationandhigher-orderconstraint-basedsegmentationoffluid-associated regionsinretinalsd-octimages,” InformationSciences,vol.358,pp. 92111,2016.
[12] ZhengY,SahniJ,CampaC,StangosAN,RajA,HardingSP.,”Computerizedassessmentofintraretinalandsubretinalfluidregionsinspectraldomainopticalcoherencetomographyimagesoftheretina,” American journalofophthalmology,vol.155,no.2,pp.277286,2013.
[13] PilchM,StiegerK,WennerY,PreisingMN,FriedburgC,zuBexten EM,etal.,”Automatedsegmentationofpathologicalcavitiesinoptical coherencetomographyscanspathologicalcavitiesinoctscans,” Investigativeophthalmology&visualscience,vol.54,no.6,pp.43854393, 2013.
[14] QuellecG,LeeK,DolejsiM,GarvinMK,AbramoffMD,Sonka M.,”Three-dimensionalanalysisofretinallayertexture:identification offluid-filledregionsinsd-octofthemacula,” IEEEtransactionson medicalimaging,vol.29,no.6,pp.13211330,2010.
[15] ChenX,NiemeijerM,ZhangL,LeeK,AbrmoffMD,SonkaM.”Threedimensionalsegmentationoffluid-associatedabnormalitiesinretinaloct:
probabilityconstrainedgraphsearch-graph-cut,” IEEEtransactionson medicalimaging,vol.31,no.8,pp.15211531,2012.
[16] RashnoA,KoozekananiDD,DraynaPM,NazariB,SadriS,Rabbani H,etal.,”Fully-automatedsegmentationoffluid/cystregionsinoptical coherencetomographyimageswithdiabeticmacularedemausingneutrosophicsetsandgraphalgorithms,” IEEETransactionsonBiomedical Engineering,2017.
[17] A.Rashno,B.Nazari,D.D.Koozekanani,P.M.Drayna,S.Sadri,H. Rabbani,K.K.Parhi,”Fully-automatedsegmentationoffluidregionsin exudativeage-relatedmaculardegenerationsubjects:Kernelgraphcut inneutrosophicdomain,” PloSone vol.12,no.10,pp.e0186949,2017.
[18] KohlerJ,RashnoA,ParhiKK,DraynaPM,RadwanS,Koozekanani DD.,”Correlationbetweeninitialvisionandvisionimprovementwith automaticallycalculatedretinalcystvolumeintreateddmeafterresolution,” InvestigativeOphthalmology&VisualScience,vol.58,no.8,pp. 953953,2017.
[19] ParhiKK,RashnoA,NazariB,SadriS,RabbaniH,DraynaPM,et al.,”Automatedfluid/cystsegmentation:Aquantitativeassessmentof diabeticmacularedema,” InvestigativeOphthalmology&VisualScience, vol.58,no.8,pp.46334633,2017.
[20] A.Rashno,K.K.Parhi,B.Nazari,S.Sadri,H.Rabbani,P.Drayna, D.D.Koozekanani,”Automatedintra-retinal,subretinalandsub-rpe cystregionssegmentationinage-relatedmaculardegeneration(amd) subjects,” InvestigativeOphthalmology&VisualScience,vol.58,no.8, pp.397397,2017.
[21] Salafian,B.,Kafieh,R.,Rashno,A.,Pourazizi,M.andSadri,S., ”Automaticsegmentationofchoroidlayerinedioctimagesusinggraph theoryinneutrosophicspace,” arXivpreprintarXiv:1812.01989,2018.
[22] Bogunovi,H.,Venhuizen,F.,Klimscha,S.,Apostolopoulos,S.,BabHadiashar,A.,Bagci,U.,Beg,M.F.,Bekalo,L.,Chen,Q.,Ciller,C. andGopinath,K.,”RETOUCH-TheRetinalOCTFluidDetectionand SegmentationBenchmarkandChallenge,” IEEEtransactionsonmedical imaging,2019.
[23] Rashno,A.,Koozekanani,D.D.andParhi,K.K.,”Octfluidsegmentation usinggraphshortestpathandconvolutionalneuralnetwork,” In2018 40thAnnualInternationalConferenceoftheIEEEEngineeringin MedicineandBiologySociety(EMBC),pp.3426-3429,2018.
[24] Montuoro,A.,Waldstein,S.M.,Gerendas,B.S.,Schmidt-Erfurth,U.and Bogunovi,H.,”JointretinallayerandfluidsegmentationinOCTscans ofeyeswithseveremacularedemausingunsupervisedrepresentation andauto-context,” Biomedicalopticsexpress,vol.8,no.3,pp.18741888,2017.
[25] Roy,A.G.,Conjeti,S.,Karri,S.P.K.,Sheet,D.,Katouzian,A., Wachinger,C.andNavab,N.,”ReLayNet:retinallayerandfluid segmentationofmacularopticalcoherencetomographyusingfully convolutionalnetworks,” Biomedicalopticsexpress,vol.8,no.8,pp. 3627-3642,2017.
[26] Smarandache,F.ed.,”Aunifyingfieldinlogics:Neutrosophiclogic: neutrosophy,neutrosophicset,neutrosophicprobability,” InfiniteStudy, 2003.
[27] Guo,Y.andCheng,H.D.,”Newneutrosophicapproachtoimage segmentation,” PatternRecognition,vol.42,no.5,pp.587-595,2009.
[28] Zhang,M.,Zhang,L.andCheng,H.D.,”Aneutrosophicapproachto imagesegmentationbasedonwatershedmethod,” SignalProcessing, vol.90,no.5,pp.1510-1517,2010.
[29] Sengur,A.andGuo,Y.,”Colortextureimagesegmentationbasedon neutrosophicsetandwavelettransformation,” ComputerVisionand ImageUnderstanding,vol.115,no.8,pp.1134-1144,2011.
[30] Heshmati,A.,Gholami,M.andRashno,A.,”Schemeforunsupervised colourtextureimagesegmentationusingneutrosophicsetandnonsubsampledcontourlettransform,” IETImageProcessing,vol.10,no. 6,pp.464-473,2016.
[31] Guo,Y.,engr,A.andYe,J.,”Anovelimagethresholdingalgorithm basedonneutrosophicsimilarityscore,” Measurement,vol.58,pp.175186,2014.
[32] Guo,Y.andengr,A.,”Anovelimageedgedetectionalgorithmbased onneutrosophicset,” Computers&ElectricalEngineering,vol.40,no. 8,pp.3-25,2014.
[33] Rashno,A.,Smarandache,F.andSadri,S.,”Refinedneutrosophicsets incontent-basedimageretrievalapplication,” In201710thIranian ConferenceonMachineVisionandImageProcessing(MVIP),pp.197202,2017.
[34] Rashno,A.andSadri,S.,”Content-basedimageretrievalwithcolor andtexturefeaturesinneutrosophicdomain,” In20173rdInternational
ConferenceonPatternRecognitionandImageAnalysis(IPRIA),pp. 50-55,2017.
[35] Rashno,A.,Sadri,S.andSadeghianNejad,H.,”Anefficientcontentbasedimageretrievalwithantcolonyoptimizationfeatureselection schemabasedonwaveletandcolorfeatures,” In2015TheInternational SymposiumonArtificialIntelligenceandSignalProcessing(AISP),pp. 59-64,2015.
[36] Rashno,A.andRashno,E.,”Content-basedimageretrievalsystemwith mostrelevantfeaturesamongwaveletandcolorfeatures,” arXivpreprint arXiv:1902.02059,2019.
[37] S.Fadaei,A.Rashno,E.Rashno,Content-basedimageretrievalspeedup, arXivpreprintarXiv:1911.11379.
[38] Rashno,E.,Akbari,A.andNasersharif,B.,”AConvoloutionalNeural NetworkmodelbasedonNeutrosophyforNoisySpeechRecognition,” InfiniteStudy,2019.
[39] Rashnoa,E.,Minaei-Bidgolia,B.andGuo,Y.,”Aneffectiveclustering methodbasedondataindeterminacyinneutrosophicsetdomain,” InfiniteStudy,2018.
[40] Guo,Y.andSengur,A.,”NCM:Neutrosophicc-meansclustering algorithm,” PatternRecognition,vol.48,no.8,pp.2710-2724,2015.
[41] Rashno,E.,Norouzi,S.S.,Minaei-Bidgoli,B.andGuo,Y.,”Certainty ofoutlierandboundarypointsprocessingindatamining,” In201927th IranianConferenceonElectricalEngineering(ICEE),pp.1929-1934, 2019.
[42] Rashno,E.andMinaei-Bidgoli,B.,”Boundarypointshandlingforimage edgedetectionbasedonNeutrosophicset,” In20195thConferenceon KnowledgeBasedEngineeringandInnovation(KBEI),pp.886-890, 2019.
[43] Bezdek,J.C.,Ehrlich,R.andFull,W.,”FCM:Thefuzzyc-means clusteringalgorithm,” Computers&Geosciences,vol.10,no.2-3,pp. 191-203,1984.
[44] Boykov,Y.andFunka-Lea,G.,”GraphcutsandefficientNDimage segmentation,” Internationaljournalofcomputervision,vol.70,no.2, pp.109-131,2006.
[45] Salah,M.B.,Mitiche,A.andAyed,I.B.,”Multiregionimagesegmentationbyparametrickernelgraphcuts,” IEEETransactionsonImage Processing,vol.20,no.2,pp.545-557,2010.
[46] Esmaeili,M.,Dehnavi,A.M.,Rabbani,H.andHajizadeh,F.,”Threedimensionalsegmentationofretinalcystsfromspectral-domainoptical coherencetomographyimagesbytheuseofthree-dimensionalcurvelet basedK-SVD,” Journalofmedicalsignalsandsensors,vol.6,no.3, p.166,2016.