DTI数据分析及应用

DTI数据分析及应用舒妮博士北京师范大学认知神经科学与学习研究所Page2内容提纲„„DTIDTI的研究内容的研究内容„„DTIDTI数据处理流程数据处理流程„„DTIStudioDTIStudio„„FSL:FSL:FMRIBFMRIB’’ssDiffusionToolboxDiffusionToolboxPage3扩散张量成像的研究内容„„纤维跟踪算法纤维跟踪算法„„基于基于DTIDTI的应用研究的应用研究Page4扩散张量的数学描述D=D=DDxxxxDDxyxyDDxzxzDDxyxyDDyyyyDDyzyzDDxzxzDDyzyzDDzzzzλλ11000000λλ22000000λλ33vv11vv22vv33==((vv11vv22vv33))特征分解特征分解特征值特征值::λλ11≥≥λλ22≥≥λλ3300特征向量特征向量::vvii⊥⊥vvjj,,ii≠≠jjD=D=DDxxxxDDxyxyDDxzxzDDDxyxyxyDDyyyyDDyzyzDDDxzxzxzDDDyzyzyzDDzzzzPage5确定性跟踪算法跟踪终止条件跟踪终止条件Morietal.,AnnNeurol,1999Page6确定性跟踪结果Catanietal,Brain,2005粗大的白质纤维束粗大的白质纤维束Page7纤维走向的不确定性Jones,MRM,2003LinearityLinearityUncertaintyUncertaintyBootstrapBootstrap方法方法Page8概率跟踪算法„„DirectionUncertaintyDirectionUncertainty‹‹DTINoiseDTINoise‹‹PartialVolumeEffectsPartialVolumeEffectsSlidefromTriNgoPage9概率跟踪的方法„„NonNon--parametric(modelfree)approachesparametric(modelfree)approaches‹‹BootstrapmethodBootstrapmethod‹‹HARDI:QHARDI:Q--ball,DSIball,DSI„„ParametricapproachesParametricapproaches‹‹Priorknowledgeandmodels:BayesianframeworkPriorknowledgeandmodels:Bayesianframework‹‹Probabilitydensityfunction(PDF):local,globalProbabilitydensityfunction(PDF):local,globalHowtoestimatethedistributionoffiberorientationswithinavoxel?Page10概率跟踪的思想Reference:Behrens,T.E.etal.Characterizationandpropagationofuncertaintyindiffusion-weightedMRimaging.MagnResonMed50,1077-88(2003).Page11概率跟踪结果Frimanetal,IEEETMI,2006Page12„„概率跟踪的优点：概率跟踪的优点：‹‹估计纤维走向的不确定性估计纤维走向的不确定性,,一定程度上解决纤一定程度上解决纤维交叉问题维交叉问题‹‹研究研究FAFA较低的灰质脑区之间的解剖连接较低的灰质脑区之间的解剖连接‹‹跟踪结果对噪声更稳定跟踪结果对噪声更稳定‹‹定量描述空间任意两个体素之间的连接概率定量描述空间任意两个体素之间的连接概率„„概率跟踪的缺点：概率跟踪的缺点：‹‹需要采集较多梯度方向的需要采集较多梯度方向的DTIDTI图像图像‹‹计算量大，耗时计算量大，耗时Page13Connectivity-basedclassificationofthalamicvoxelsproducesclustersBehrensetal,NatureNeuroscience,2003Page14ImprovementsonthediffusiontensormodelsinglefibremultiplefibresSlidefromSaadJbabdiPage15„„确定性跟踪常用软件确定性跟踪常用软件::‹‹DTIStudio,DTIStudio,MedINRIAMedINRIA,3DSlicer,3DSlicer等等„„概率跟踪常用软件：概率跟踪常用软件：‹‹FSLFSL扩散张量成像的研究内容„„纤维跟踪算法纤维跟踪算法„„基于基于DTIDTI的应用研究的应用研究Page17扩散属性测度以上三种情况的ADC=0.7x10-3mm2/sPage18Page19Page20Page21基于扩散属性测度的临床研究„„基于全脑配准的分析方法基于全脑配准的分析方法‹‹基于体素的统计分析（基于体素的统计分析（VBAVBA））‹‹基于白质骨架的空间统计分析（基于白质骨架的空间统计分析（TBSSTBSS））„„基于感兴趣区的分析方法基于感兴趣区的分析方法‹‹手工画感兴趣区的方法手工画感兴趣区的方法‹‹基于纤维重建的定量分析基于纤维重建的定量分析Page22Voxel-BasedAnalysis(VBA)„„VBMonFA(VBMonFA(AshburnerAshburner,2000;,2000;RuggRugg--Gunn,2001)Gunn,2001)„„StrengthsStrengths¾¾Fullyautomated&quickFullyautomated&quick¾¾InvestigationwholebrainInvestigationwholebrain„„ImplementationstepsImplementationsteps¾¾PreprocessingPreprocessing¾¾NormalizationofFAimagesNormalizationofFAimages¾¾SmoothSmooth¾¾VoxelwiseVoxelwisestatistics(e.g.controlsstatistics(e.g.controls––patients)patients)„„IssuesIssues¾¾Alignmentdifficult;smallestsystematicshiftsbetweenAlignmentdifficult;smallestsystematicshiftsbetweengroupscanbeincorrectlyinterpretedasFAchangegroupscanbeincorrectlyinterpretedasFAchange¾¾Noobjectivewaytochoosesmoothingextent(6,8or10Noobjectivewaytochoosesmoothingextent(6,8or10mm?)mm?)Page23Page24Page25Page26Page27Page28Page29Page30Page31精神分裂症患者的VBA分析FAFA降低的脑区：降低的脑区：•cerebralpeduncle;•frontalregions;•inferiortemporalgyrus;•medialparietallobes;•hippocampalgyrus;•Insula;•rightanteriorcingulumbundle;•rightcoronaradiataHaoetal.Neuroreport2006Page32Tract-BasedSpatialStatistics(TBSS)„„PartofFSLsoftwarePartofFSLsoftware()„„OvercomethedrawbacksinVBAmethod,suchasOvercomethedrawbacksinVBAmethod,suchasalignmentissueandsmoothingissuealignmentissueandsmoothingissue„„FlowchartFlowchartPage33„„TBSSstepsindetail:TBSSstepsindetail:‹‹preprocessingpreprocessing--createFAimagesfromyourdiffusioncreateFAimagesfromyourdiffusionstudydatastudydata‹‹tbss_1_preproctbss_1_preproc--prepareyourFAdatainyourTBSSprepareyourFAdatainyourTBSSworkingdirectoryintherightformatworkingdirectoryintherightformat‹‹tbss_2_regtbss_2_reg--applynonlinearregistrationofallFAapplynonlinearregistrationofallFAimagesintostandardspaceimagesintostandardspace‹‹tbss_3_postregtbss_3_postreg--createthemeanFAimageandcreatethemeanFAimageandskeletoniseskeletoniseitit‹‹tbss_4_prestatstbss_4_prestats--projectallsubjects'FAdataontotheprojectallsubjects'FAdataontothemeanFAskeletonmeanFAskeleton‹‹statsstats(e.g.,(e.g.,randomiserandomise))--feedthe4DprojectedFAdatafeedthe4DprojectedFAdataintoGLMintoGLMmodellingmodellingandandthresholdingthresholdinginordertofindinordertofindvoxelsvoxelswhichcorrelatewithyourmodel.whichcorrelatewithyourmodel.Page34Docross-subjectvoxelwisestatsonskeleton-projectedFAPage35Fig.TBSSresultsfrom15MSpatients.A,B:3DsurfacerenderingsofthemeanFAskeleton.C:YellowshowsthewhereFAcorrelatesnegativelywithEDSSdisabilityscore.D:Redasabove.InCandD,greenshowsthemeanFAskeleton,blueshowsthegroupmeanlesiondistribution,andthebackgroundimageistheMNI152.Smithetal.,NeuroImage,2006Page36ScholzScholzetal.,NatureNeuroscience2009etal.,NatureNeuroscience2009Page37„„TBSSdataacquisitionrequirement:TBSSdataacquisitionrequirement:‹‹VoxelVoxelsize