《DTI数据分析及应用.ppt》由会员分享，可在线阅读，更多相关《DTI数据分析及应用.ppt（74页珍藏版）》请在三一办公上搜索。
1、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,Page 2,内容提纲,DTI的研究内容,DTI数据处理流程,DTI Studio,FSL:FMRIBs Diffusion Toolbox,Page 3,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,Dxyxy Dyy Dyz=(v1 v2 v3)0 2 0,扩散张量的数学描述,D=,特征分解特征值:123 0特征向量:vivj,i jPage 4,Dxx Dxy Dxz 1 0 0 v1v2Dxz Dyz Dzz 0 0 3 v3,Page 5,确定性跟踪算法,跟踪终止条件,Mori et al.,An
2、n Neurol,1999,Page 6,确定性跟踪结果,Catani et al,Brain,2005,粗大的白质纤维束,Uncertainty,Page 7,纤维走向的不确定性,Jones,MRM,2003,Linearity,Bootstrap 方法,Page 8,概率跟踪算法,Direction Uncertainty,DTI Noise,Partial Volume Effects,Slide from Tri Ngo,Page 9,概率跟踪的方法,Non-parametric(model free)approaches,Bootstrap method,HARDI:Q-ball,D
3、SI,Parametric approaches,Prior knowledge and models:Bayesian frameworkProbability density function(PDF):local,global,How to estimate the distribution of fiber,orientations within a voxel?,Page 10,概率跟踪的思想Reference:Behrens,T.E.et al.Characterization and propagation of uncertainty,in diffusion-weighted
4、 MR imaging.Magn Reson Med 50,1077-88(2003).,Page 11,概率跟踪结果,Friman et al,IEEE TMI,2006,Page 12,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,Page 13,Connectivity-based classification ofthalamic voxels produces clusters,Behrens et
5、 al,Nature Neuroscience,2003,Page 14,Improvements on the diffusion tensor model,single fibre,multiple fibres,Slide from Saad Jbabdi,Page 15,确定性跟踪常用软件:,DTI Studio,MedINRIA,3D Slicer等,概率跟踪常用软件：,FSL,Page 16,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,Page 17,扩散属性测度,以上三种情况的 ADC=0.7 x 10-3 mm2/s,Page 18,Page 19,Page 2
6、0,Page 21,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,Page 22,Voxel-Based Analysis(VBA),VBM on FA(Ashburner,2000;Rugg-Gunn,2001)Strengths,Fully automated&quickInvestigation whole brain,Implementation steps,Preprocessing,Normalization of FA imagesSmo
7、oth,Voxelwise statistics(e.g.controls patients),Issues,Alignment difficult;smallest systematic shifts betweengroups can be incorrectly interpreted as FA change,No objective way to choose smoothing extent(6,8 or 10mm?),Page 23,Page 24,Page 25,Page 26,Page 27,Page 28,Page 29,Page 30,精神分裂症患者的VBA分析FA降低的
8、脑区：,cerebral peduncle;frontal regions;inferior temporalgyrus;medial parietal lobes;hippocampal gyrus;Insula;right anteriorcingulum bundle;right corona radiataPage 31,Hao et al.Neuroreport 2006,Page 32,Tract-Based Spatial Statistics,(TBSS),Part of FSL software,(http:/www.fmrib.ox.ac.uk/fsl/tbss/index
9、.html),Overcome the drawbacks in VBA method,such asalignment issue and smoothing issueFlowchart,Page 33,TBSS steps in detail:,preprocessing-create FA images from your diffusionstudy data,tbss_1_preproc-prepare your FA data in your TBSSworking directory in the right format,tbss_2_reg-apply nonlinear
10、registration of all FAimages into standard space,tbss_3_postreg-create the mean FA image andskeletonise it,tbss_4_prestats-project all subjects FA data onto themean FA skeleton,stats(e.g.,randomise)-feed the 4D projected FA datainto GLM modelling and thresholding in order to findvoxels which correla
11、te with your model.,Page 34,Do cross-subject voxelwise stats on,skeleton-projected FA,Page 35,Fig.TBSS results from15 MS patients.A,B:3D surface renderingsof the mean FAskeleton.C:Yellowshows the where FAcorrelates negativelywith EDSS disabilityscore.D:Red as above.In C and D,greenshows the mean FAs
12、keleton,blue showsthe group mean lesiondistribution,and thebackground image isthe MNI152.,Smith et al.,NeuroImage,2006,Page 36,Scholz et al.,Nature Neuroscience 2009,Page 37,TBSS data acquisition requirement:Voxel size should be less than 3 3 3 mm3.,At least one b=0 should be acquired;ideally one b=
13、0image for every eight diffusion-weighted images.,b-value should be at least 800 s/mm2.,At least six-gradient directions must be acquired.it isbetter to use more unique sampling directions(withisotropic angular density18)than to obtain repeatsamples of the same set of directions.,SNR in the diffusio
14、n-weighted images should bemaximized.An example protocol that should lead to,sufficiently high SNR is having b=1,000 s mm2,24,diffusion-weighted images and SNR greater than 15 inthe b=0 image.,Page 38,The data should not be upsampled(e.g.,through unfilteredzero-padding during reconstruction)if this
15、is done in such away as to introduce ringing into the data.,If multiple repeats of b=0 or diffusion-weighted images are tobe acquired,they must not be averaged on the scanner(astheymust be coregistered before averaging,and any risk ofaveraging the complex data should be avoided).,Fat saturation shou
16、ld be used whenever possible to removesignal from the scalp,which can disrupt signal in the brainowing to chemical shift or ghosting artifacts.,A vitamin capsule leftright marker(oil,not water)should beattached to the right side of the head to avoid any leftrightambiguities during data conversion an
17、d analysis.,Page 39,Computing equipment:,Unix-based computers.AppleMac(running Mac OS Xversion 10.4 or higher)and PCs(running Linux flavorsRedHat 9,Enterprise,FC4,Suse 9.0-9.3 or Debian),High RAM requirements(particularly if tens ofsubjects are used in a study),it is likely that the,computer will ne
18、ed to be 64 bit.The computer shouldhave at least a 1 GHz CPU clock,1 GB RAM,5 GBswap and 20 GB free hard disk space.,If multiple networked computers(or a computer cluster)are available,the registration steps can be parallelized,greatly reducing the total computation time.,Page 40,白质纤维束的定量分析,FA:Left
19、Cingulum(Red)Right Cingulum(Blue),Parameterization process,Gong et al,Human Brain Mapping,2005,Page 41,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23显著升高。,早期盲人大脑白质扩散异常研究,Shu et al,Human Brain Mapping,2009,Page 42,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,Page 43,内容提纲,DTI的研究内容,DTI数据处理流程,DTI Studio,FSL
20、:FMRIBs Diffusion Toolbox,Page 44,数据处理的基本流程,Page 45,DWI from ScannerS0S1S2S3S4S5,S6,Page 46,Preprocessing,DICOM data conversion,Image quality check,Eddy current correction,Page 47,内容提纲,DTI的研究内容,DTI数据处理流程,DTI Studio,FSL:FMRIBs Diffusion Toolbox,Page 48,DTI Studio,Download&InstallUser ManualMailing li
21、st,Page 49,Launching the Program and HardwareRequirement,DtiStudio-latest-x86.exe for Windowssystem,More than 1GB RAM is recommended,Page 50,Main Functions,Image Viewer,Diffusion Tensor CalculationFiber Tracking and Editing3D Visualization,Image File Management,ROI Drawing and Statistics,Page 51,How
22、 to do tensor calculation,and fiber tracking?,Page 52,E:workTrainingExampleData,Raw data:MRIcroN dcm2nii.exe,(.img,.hdr,.bvec,.bval),Eddy current correction:AIR,Tensor,FA,MD calculation:DTIstudioFiber tractography:DTIstudioROI selection,Page 53,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
23、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（Output Format选择4D NIfTI hdr/img）。,Page 54,第二步：对DTI图像进行头动和涡流校正。打开DTI studio,File-MRI View3D,选中上一步得到的4D.img文件，Image Parameters中选择Image File Format为Analyze，点击OK，然后在Image面板Image Processing区域选择AutomaticImage Registration(AIR)，按图3进行设置，然后点击OK，等图像配准完成后，在
24、Image面板的Orthogonal Views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Raw Data，保存为一个4D的.dat文件。,MRI View,3D参数,Page 55,AIR的参数设置,Page 56,头动和涡流校正后的DTI图像保存,Page 57,第三步：张量解算以及FA,ADC等扩散指标的计算。打开DTI studio,File-DTI Mapping,选择Philips REC格式，Continue，按图5进行参数设置，Add a file中选中上一步保存的4D.dat文件，点击OK，在,D
25、tiMap面板的Calculation区域选中Tensor,Color Map etc.（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Orthogonal View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。,Page 58,DtiMap面板进行张量解算,Page 59,各种扩散属性的显示,Page 60,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Fiber Tra
26、cking区域点击Fiber Tracking，然后进行参数设置，点击OK，就会进行基于全脑体素（FA0.2）的纤维重建；第二种方法：如果上一步已保存FA和Eigen Vector-0文件，可重新打开DTI Studio,File-Fiber-Tracking，选上FA-Map文件和Principle Vector文件，并进行参数设置，点击OK，就会进行基于全脑体素（FA0.2）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。,Page 61,纤维跟踪方法2,Page 62,纤维跟踪的参数设置,Page 6
27、3,重建纤维束的可视化,Page 64,Page 65,Major white matter tracts,Reference:Wakana S,Caprihan A,Panzenboeck MM,et al.,Reproducibility ofquantitative tractography methods applied to cerebral white matter.Neuroimage,2007,36(3):630-644.,Page 66,纤维属性的统计分析,Page 67,New Modules,ROIEditor,ROI drawing,selection and mani
28、pulationUser defined ROI statistics,Atlas&WMPM based statistics,Tract specific probabilistic quantification,DiffeoMap,Image registration,Page 68,内容提纲,DTI的研究内容,DTI数据处理流程,DTI Studio,FSL:FMRIBs Diffusion Toolbox,Page 69,FSL,FDT:FMRIBs Diffusion Toolbox-DWIAnalysis and Tractography,eddy_correctdtifit,be
29、dpostx,probtrackx,DTI data with 30 gradient directions,Page 70,Page 71,Processing steps:,cd/fsl_course_data/fdt/subj1dcm2nii,Bring up fslview and open data,Extracting nodif and generating a brain maskCorrecting for eddy current and head motionDtifit-fitting diffusion tensors to the dataBedpostx(2530h/person)Probtrackx,Page 72,Commands:,eddy_correct sub001.nii data 0fslroi data nodif 0 1,bet nodif nodif_mask-m-f 0.3,dtifit-k data-m nodif_brain_mask-rbvecs-b bvals-o dti,bedpostx,可参考,Page 73,Reference Book,Email:,Page 74,谢谢大家！,
