基于自适应窗口化方法小波变换的的数字化乳房X线照片肿瘤检测(IJMECS-V4-N3-8)

I.J.ModernEducationandComputerScience,2012,3,57-65PublishedOnlineApril2012inMECS()DOI:10.5815/ijmecs.2012.03.08Copyright©2012MECSI.J.ModernEducationandComputerScience,2012,3,57-65DetectionofTumoursinDigitalMammogramsUsingWaveletBasedAdaptiveWindowingMethodG.BharathaSreejaPGCommunicationSystems,CapeInstituteofTechnology,Levengipuram,IndiaEmail:[e-mail],ECEDept.,CapeInstituteofTechnology,Levengipuram,IndiaEmail:[e-mail],R&D,KalasalingamUniversity,Krishnankoil,IndiaEmail:[e-mail]—Mammographyisthemosteffectiveprocedurefortheearlydetectionofbreastdiseases.Mammogramanalysisreferstheprocessingofmammogramswiththegoaloffindingabnormalitypresentedinthemammogram.Inthispaper,thetumourcanbedetectedbyusingwaveletbasedadaptivewindowingtechnique.Coarsesegmentationisthefirststepwhichcanbedonebyusingwaveletbasedhistogramthresholdingwhere,thetheresholdvalueischosenbyperforming1-DwaveletbasedanalysisofPDFsofwavelettransformedimagesatdifferentchannels.Finesegmentationcanbedonebypartitioningtheimageintofixednumberoflargeandsmallwindows.Bycalculatingthemean,maximumandminimumpixelvaluesforthewindowsathresholdvaluehasbeenobtained.Dependinguponthethresholdvaluesthesuspiciousareashavebeensegmented.Intensityadjustmentisappliedasapreprocessingsteptoimprovethequalityofanimagebeforeapplyingtheproposedtechnique.ThealgorithmisvalidatedwithmammogramsinMammographicImageAnalysisSocietyMiniMammographicdatabasewhichshowsthattheproposedtechniqueiscapableofdetectinglesionsofverydifferentsizes.IndexTerms—waveletbasedThresholding,breastcancer,mammography,windowbasedThresholding,segmentation.I.INTRODUCTIONCurrently,breastcancerisaleadingcauseofdeathamongwomenandsecondmajorcauseofdeathafterlungcancer[1]-[5].BreastcanceristhesecondmostcommoncancerinIndianwomen.Theincidenceismoreinurbanthanruralwomen.Itismoreprevalentinthehighersocio-economicgroups.Theaverageincidenceratevariesfrom22-28per1,00,000womenperyearinurbansettingsto6per100,000womenperyearinruralareas.Duetorapidurbanizationandwesternizationoflifestyles,thereisarisingincidenceofbreastcancerinIndia.AccordingtoTheInternationalAgencyforResearchonCancer,whichispartoftheWorldHealthOrganization,therewereapproximately79,000womenperyearaffectedbybreastcancerinIndia.Itisthoughtthatittakesabout10yearsforatumourtobecome1cminsizestartingfromasinglecell.Earlierdiagnosesofbreastcancerareofgreatimportanceinmodernmedicine.Atpresent,mammographyisthemethodofchoiceforearlybreastcancerdetection[6]-[8].Althoughautomaticanalysisofmammogramscannotfullyreplaceradiologists,anaccuratecomputer-aidedanalysismethodcanhelpradiologiststomakemorereliableandefficientdecisions[9].Tumorsandotherabnormalitiespresentinthemammogramsareofbasicintereststhatneedtobesegmentedandextractedinmammograms[10]-[11].Someofthegrayscalebasedsegmentationmethodsarequiteeffectivetoextracttheexactedgesofhomogeneousgrayscaleregions.Theyareoftennotsoeffectivetoextractthedesiredaffectedareasinmammogramswithcomplexstructurebecauseofthecomplexdistributionofthegrayscale.However,theappearancesofbreastcancersareverysubtleandunstableintheirearlystages.Therefore,doctorsandradiologistscanmisstheabnormalityeasilyiftheyonlydiagnosebyexperience.Thecomputeraideddetectiontechnologycanhelpdoctorsandradiologistsingettingamorereliableandeffectivediagnosis.Therearenumeroustumourdetectiontechniqueshave58DetectionofTumoursinDigitalMammogramsUsingWaveletBasedAdaptiveWindowingMethodCopyright©2012MECSI.J.ModernEducationandComputerScience,2012,3,57-65beenanalyzed[12]-[13].Aregion-growingtechniquehasbeendiscussedbyUmeshAdigaetal[14]whichisconstrainedbyshapeandsizesimilaritiesofcellnuclei.Inhighthroughputtissueimageanalysis,thismethodrequiresanautomaticinitialseedselectionandbecomesslowduetotherequirementofcontinuousupdatingofsimilaritymeasures.Thesameissueslimitthesemi-automationmethodproposedbyWuandBarba[15].Whenmorphologicalfiltersareused,theshapeofthesegmentedobjecttendstochangeinaccordancewiththeshapeofthestructuringelementsusedforfiltering[16]-[18].Wuetal.proposedaparametricmodel-fittingalgorithmforcellsegmentation[19].Inthismethod,theyassumeobjectsareconvexandhenceashapemodelcanbeusedforsegmentation.Thecompositionofthealgorithmisalsoverycomplicatedwhenoverlappedstructurearepresentintheimage.Wavelettransform-basedmethodsofferanaturalframeworkforprovidingmultiscaleimagerepresentationsthatcanbeseparatelyanalyzed[20]-[25].Throughamultiscaledecomposition[26]-[28],mostofthegrossintensitydistributioncanbeisolatedinalargescaleimage,whiletheinformationaboutdetailsandsingularities,suchasedgesandtextures,canbeisolatedinmid-tosmallscales.Here1-Dwavelet-basedanalysisisperformedtofindthePDFandadaptivelyselectedproperthresholdsforsegmentationbysearchingforthelocalminimaofthe1-DwavelettransformedPDF[29].Thismethodissimple,fast,andeffectiveforsegmentingtumorsinmammograms.However,themethodisnotveryeffectivewhenthetargetandthebackgroundregionsdemonstratelittledifferenceing