資源描述:
《人體腦圖像分割技術(shù)研究》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫(kù)。
1、摘要摘要醫(yī)學(xué)圖像分割是醫(yī)學(xué)圖像處理和分析領(lǐng)域的基礎(chǔ)性經(jīng)典難題,其中腦部醫(yī)學(xué)圖像分割因其重要的應(yīng)用價(jià)值一直是醫(yī)學(xué)圖像分割的研究熱點(diǎn)。本論文的主要工作是對(duì)可視人腦部圖像和MRI腦部圖像進(jìn)行分割技術(shù)研究,以便提取出特定n0組織和結(jié)構(gòu)。論文的創(chuàng)新點(diǎn)包括:1.提出了一種基于最大熵和數(shù)學(xué)形態(tài)學(xué)的可視人腦部彩色圖像分割方法。該方法是將原本只用于灰度圖像分割的最大熵方法和數(shù)學(xué)形態(tài)學(xué)方法有機(jī)結(jié)合而形成的一種彩色腦圖像分割方法,它不僅能對(duì)可視人腦部圖像進(jìn)行有效分割,斯凡成功提取出了腦組織和腦骨結(jié)構(gòu)。經(jīng)對(duì)多達(dá)576幅可視人腩部圖像作實(shí)驗(yàn)均獲社}若用單一算法無(wú)法實(shí)現(xiàn)的滿意效果。2.為適應(yīng)MRI圖象分割
2、的問(wèn)題提出了種新的核聚類算法。該方法針對(duì)傳統(tǒng)模糊核聚類算法當(dāng)數(shù)據(jù)類差別很大時(shí),小數(shù)據(jù)類被誤分或被大數(shù)據(jù)類吞并的缺陷,通過(guò)定義一個(gè)新的目標(biāo)函數(shù),為每一個(gè)類分配了一個(gè)動(dòng)態(tài)權(quán)值,可改善聚類效果,實(shí)驗(yàn)結(jié)果表明,本文算法對(duì)腦白質(zhì)、腦灰質(zhì)、腦脊液分割結(jié)果在降低誤分率性能方面均比經(jīng)典核聚類算法有5%以上的提高。3.提出了一種分層Mumford.Shah模型的顱腦圖像分割方法。該方法針。劉1Mumford.Shah模型分割多目標(biāo)時(shí)初始化的耦合對(duì)分割結(jié)果有明顯影響的問(wèn)題,提出應(yīng)用分層Mumford—Shah模型進(jìn)行圖像分割,并運(yùn)用了改進(jìn)的圖像分割方羈;進(jìn)行計(jì)算求解的圖像分割方法,經(jīng)對(duì)MRI腦圖像
3、腦白質(zhì),臟i狄質(zhì)和腦脊液的分割實(shí)驗(yàn),取得了良好的效果。關(guān)鍵詞:醫(yī)學(xué)圖像分割最大熵?cái)?shù)學(xué)形態(tài)學(xué)模糊c均值聚類水平集Mumford.Shah模型AbstractMedicalimagesegmentationisahard—toughprobleminmedicalimageprocessingandanalysis.Amongit,brainmedicalimagesegmentationistheresearchfocusforitsimportantvalues.TheprimarytaskofnaythesisisstndyingtheVis.ibleHumanbrainin
4、aagesegmentationandMRIbrainimagesegmentationtechniquesinordertoextractthetissuesandstructures.Theachievementsofmythesisinclude:IAVisibleHumanbraincolorimagesegmentationmethodbasedonmaxilltumentropyhasbeenproposed.CombingthemaximumentropymethodandMathematicalMorphologymethodthatwereonlyusedfo
5、rgrayimageoriginally.OLIFmethodextractsbraintissuesandbrainbonesthatwon’tbedoneifsinglemethodisused.Theexamplesof576brainimagesshowitseffectiveness.2.AnewMRIkernelfuzzyclusteringalgorithmhasbeenproposedforMRIimagesegmentation,Whenthedifferenceofthedatesizeislarge.Thesmallclassmaybemisclassif
6、iedormaybemergedintobigclassinfuzzykernelclusteringalgorithm.Anewobjectivefunctionisintroducedandanadditionalweightingthctorisassignedtoeachclass.Basedonthisimprovementthealgoritlungetsbetterperformance.7l'heresultsindicateitcouldreduce5%misclassificationrateofthewhitematter、graymatterandCSF
7、thanoriginalalgorithm.3.WhenMumford—Shahmodelisusedformulti-objectsegmentation.theinitializationinfluencestheresultsseriously.Anovelimagesegmentationapproachbased01lhierarchicalMumford—Shahmodelhasbeenproposedinthispaper,whichgetsthesolutionbysolvi