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DTI的研究内容DTI数据处理流程DTIStudioFSL:FMRIB’sDiffusionToolboxPage3扩散张量成像的研究内容纤维跟踪算法基于DTI的应用研究DxyxyDyyDyz=(v1v2v3)0λ20扩散张量的 数学 数学高考答题卡模板高考数学答题卡模板三年级数学混合运算测试卷数学作业设计案例新人教版八年级上数学教学计划 描述D= 特征分解特征值:λ1≥λ2≥λ3>0特征向量:vi⊥vj,i≠j Page4DxxDxyDxzλ100v1 v2DxzDyzDzz00λ3v3Page5确定性跟踪算法跟踪终止条件Morietal.,AnnNeurol,1999Page6确定性跟踪结果Catanietal,Brain,2005粗大的白质纤维束UncertaintyPage7纤维走向的不确定性Jones,MRM,2003LinearityBootstrap方法Page8概率跟踪算法DirectionUncertaintyDTINoisePartialVolumeEffectsSlidefromTriNgoPage9概率跟踪的方法Non-parametric(modelfree)approachesBootstrapmethodHARDI:Q-ball,DSIParametricapproachesPriorknowledgeandmodels:BayesianframeworkProbabilitydensityfunction(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：估计纤维走向的不确定性,一定程度上解决纤维交叉问题研究FA较低的灰质脑区之间的解剖连接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概率跟踪的缺点：需要采集较多梯度方向的DTI图像计算量大，耗时Page13Connectivity-basedclassificationof thalamicvoxelsproducesclustersBehrensetal,NatureNeuroscience,2003Page14ImprovementsonthediffusiontensormodelsinglefibremultiplefibresSlidefromSaadJbabdiPage15确定性跟踪常用软件:DTIStudio,MedINRIA,3DSlicer等概率跟踪常用软件：FSLhttp://www.fmrib.ox.ac.uk/fsl/fdt/Page16扩散张量成像的研究内容纤维跟踪算法基于DTI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体素的统计分析（VBA）基于白质骨架的空间统计分析（TBSS）基于感兴趣区的分析方法手工画感兴趣区的方法基于纤维重建的定量分析Page22Voxel-BasedAnalysis(VBA)VBMonFA(Ashburner,2000;Rugg-Gunn,2001)StrengthsFullyautomated&quickInvestigationwholebrainImplementationstepsPreprocessingNormalizationofFAimagesSmoothVoxelwisestatistics(e.g.controls–patients)IssuesAlignmentdifficult;smallestsystematicshiftsbetweengroupscanbeincorrectlyinterpretedasFAchangeNoobjectivewaytochoosesmoothingextent(6,8or10mm?)Page23Page24Page25Page26Page27Page28Page29Page30精神分裂症患者的VBA分析 FA降低的脑区：••••••••cerebralpeduncle;frontalregions;inferiortemporalgyrus;medialparietallobes;hippocampalgyrus;Insula;rightanteriorcingulumbundle;rightcoronaradiata Page31Haoetal.Neuroreport2006Page32Tract-BasedSpatialStatistics(TBSS)PartofFSLsoftware(http://www.fmrib.ox.ac.uk/fsl/tbss/index.html)OvercomethedrawbacksinVBAmethod,suchasalignmentissueandsmoothingissueFlowchartPage33TBSSstepsindetail:preprocessing-createFAimagesfromyourdiffusionstudydatatbss_1_preproc-prepareyourFAdatainyourTBSSworkingdirectoryintherightformattbss_2_reg-applynonlinearregistrationofallFAimagesintostandardspacetbss_3_postreg-createthemeanFAimageandskeletoniseittbss_4_prestats-projectallsubjects'FAdataontothemeanFAskeletonstats(e.g.,randomise)-feedthe4DprojectedFAdataintoGLMmodellingandthresholdinginordertofindvoxelswhichcorrelatewithyourmodel.Page34Docross-subjectvoxelwisestatsonskeleton-projectedFAPage35Fig.TBSSresultsfrom15MSpatients.A,B:3DsurfacerenderingsofthemeanFAskeleton.C:YellowshowsthewhereFAcorrelatesnegativelywithEDSSdisabilityscore.D:Redasabove.InCandD,greenshowsthemeanFAskeleton,blueshowsthegroupmeanlesiondistribution,andthebackgroundimageistheMNI152.Smithetal.,NeuroImage,2006Page36Scholzetal.,NatureNeuroscience2009Page37TBSSdataacquisitionrequirement: Voxelsizeshouldbelessthan3×3×3mm3.Atleastoneb=0shouldbeacquired;ideallyoneb=0imageforeveryeightdiffusion-weightedimages.b-valueshouldbeatleast800s/mm2.Atleastsix-gradientdirectionsmustbeacquired.itisbettertousemoreuniquesamplingdirections(withisotropicangulardensity18)thantoobtainrepeatsamplesofthesamesetofdirections.SNRinthediffusion-weightedimagesshouldbemaximized.AnexampleprotocolthatshouldleadtosufficientlyhighSNRishavingb=1,000smm–2,24diffusion-weightedimagesandSNRgreaterthan15intheb=0image.Page38Thedatashouldnotbeupsampled(e.g.,throughunfilteredzero-paddingduringreconstruction)ifthisisdoneinsuchawayastointroduceringingintothedata.Ifmultiplerepeatsofb=0ordiffusion-weightedimagesaretobeacquired,theymustnotbeaveragedonthescanner(astheymustbecoregisteredbeforeaveraging,andanyriskofaveragingthecomplexdatashouldbeavoided).Fatsaturationshouldbeusedwheneverpossibletoremovesignalfromthescalp,whichcandisruptsignalinthebrainowingtochemicalshiftorghostingartifacts.Avitamincapsuleleft–rightmarker(oil,notwater)shouldbeattachedtotherightsideoftheheadtoavoidanyleft–rightambiguitiesduringdataconversionandanalysis.Page39Computingequipment:Unix-basedcomputers.AppleMac(runningMacOSXversion10.4orhigher)andPCs(runningLinuxflavorsRedHat9,Enterprise,FC4,Suse9.0-9.3orDebian)HighRAMrequirements(particularlyiftensofsubjectsareusedinastudy),itislikelythatthecomputerwillneedtobe64bit.Thecomputershouldhaveatleasta1GHzCPUclock,1GBRAM,5GBswapand20GBfreeharddiskspace.Ifmultiplenetworkedcomputers(oracomputercluster)areavailable,theregistrationstepscanbeparallelized,greatlyreducingthetotalcomputationtime.Page40白质纤维束的定量分析FA:LeftCingulum(Red)>RightCingulum(Blue)ParameterizationprocessGongetal,HumanBrainMapping,2005Page41同正常人相比，早期盲人的视放射白质扩散异常，FA值显著降低，ADC和λ23显著升高。早期盲人大脑白质扩散异常研究Shuetal,HumanBrainMapping,2009Page42单张量模型的假设无法解决纤维交叉问题纤维跟踪技术的准确性缺乏严格的评价体系扩散张量成像的局限性Page43内容提纲DTI的研究内容DTI数据处理流程DTIStudioFSL:FMRIB’sDiffusionToolboxPage44数据处理的基本流程Page45 DWIfromScannerS0 S1 S2 S3 S4 S5S6……Page46PreprocessingDICOMdataconversionImagequalitycheckEddycurrentcorrectionPage47内容提纲DTI的研究内容DTI数据处理流程DTIStudioFSL:FMRIB’sDiffusionToolboxPage48DTIStudiohttps://www.mristudio.org/Download&InstallUserManualMailinglistPage49LaunchingtheProgramandHardwareRequirementDtiStudio-latest-x86.exeforWindowssystemMorethan1GBRAMisrecommendedPage50MainFunctionsImageViewerDiffusionTensorCalculationFiberTrackingandEditing3DVisualizationImageFileManagementROIDrawingandStatisticsPage51Howtodotensorcalculationandfibertracking?Page52E:\work\Training\ExampleDataRawdata:MRIcroNdcm2nii.exe(.img,.hdr,.bvec,.bval)Eddycurrentcorrection:AIRTensor,FA,MDcalculation:DTIstudioFibertractography:DTIstudioROIselectionPage53第一步：对原始DICOM数据进行格式转换。利用MRIcroN软件中的dcm2nii.exe工具，将DTI原始数据文件夹拖入，即可得到DTI扫描的梯度编码文件.bval和.bvec，以及转换后的NIFTI格式的图像文件（OutputFormat选择4DNIfTIhdr/img）。Page54第二步：对DTI图像进行头动和涡流校正。打开DTIstudio,File->MRIView3D,选中上一步得到的4D.img文件，ImageParameters中选择ImageFileFormat为Analyze，点击OK，然后在Image面板ImageProcessing区域选择AutomaticImageRegistration(AIR)，按图3进行设置，然后点击OK，等图像配准完成后，在Image面板的OrthogonalViews区域的文件下拉框中看到Air_开头的一系列文件，为校正后的DTI图像文件，点击Save，将Air_开头的所有文件选中，选择RawData，保存为一个4D的.dat文件。MRIView3D参数Page55AIR的参数设置Page56头动和涡流校正后的DTI图像保存Page57第三步：张量解算以及FA,ADC等扩散指标的计算。打开DTIstudio,File->DTIMapping,选择PhilipsREC格式，Continue，按图5进行参数设置，Addafile中选中上一步保存的4D.dat文件，点击OK，在DtiMap面板的Calculation区域选中Tensor,ColorMapetc.（计算ADC值选择ADC-Map），根据图像选择噪声水平，点击OK，然后等DTIStudio算完后在Image面板的OrthogonalView区域可看到计算出来的各种扩散属性文件。对于想要保存的文件，如FA,EigenVector-0，ColorMap-0等可以分别进行Save（.dat格式），便于下一次查看和使用。Page58DtiMap面板进行张量解算Page59各种扩散属性的显示Page60第四步：纤维跟踪及可视化。第一种方法：基于前面步骤，在DtiMap面板的FiberTracking区域点击FiberTracking，然后进行参数设置，点击OK，就会进行基于全脑体素（FA>0.2）的纤维重建；第二种方法：如果上一步已保存FA和EigenVector-0文件，可重新打开DTIStudio,File->Fiber-Tracking，选上FA-Map文件和PrincipleVector文件，并进行参数设置，点击OK，就会进行基于全脑体素（FA>0.2）的纤维重建。通过任何一种方法，算完后右下角会出现Fiber面板，再此面板中可以对特定纤维束进行显示和编辑，并可以对纤维属性进行统计分析。Page61纤维跟踪方法2Page62纤维跟踪的参数设置Page63重建纤维束的可视化Page64Page65MajorwhitemattertractsReference:WakanaS,CaprihanA,PanzenboeckMM,etal.,Reproducibilityofquantitativetractographymethodsappliedtocerebralwhitematter.Neuroimage,2007,36(3):630-644.Page66纤维属性的统计分析Page67NewModulesROIEditorROIdrawing,selectionandmanipulationUserdefinedROIstatisticsAtlas&WMPMbasedstatisticsTractspecificprobabilisticquantificationDiffeoMapImageregistrationPage68内容提纲DTI的研究内容DTI数据处理流程DTIStudioFSL:FMRIB’sDiffusionToolboxPage69FSLFDT:FMRIB'sDiffusionToolbox-DWIAnalysisandTractographyeddy_correctdtifitbedpostxprobtrackxDTIdatawith≥30gradientdirectionsPage70Page71Processingsteps:cd~/fsl_course_data/fdt/subj1dcm2niiBringupfslviewandopendataExtractingnodifandgeneratingabrainmaskCorrectingforeddycurrentandheadmotionDtifit-fittingdiffusiontensorstothedataBedpostx(25~30h/person)ProbtrackxPage72Commands:eddy_correctsub001.niidata0fslroidatanodif01betnodifnodif_mask-m-f0.3dtifit-kdata-mnodif_brain_mask-rbvecs-bbvals-odtibedpostx<subjectdirectory>可参考http://www.fmrib.ox.ac.uk/fsl/fdt/index.htmlPage73ReferenceBookEmail:nshu55@gmail.comPage74谢谢大家！