fMRI-SPM原理與應(yīng)用課件

StatisticalParametricMapping基本原理與使用北京師范大學(xué)認(rèn)知神經(jīng)科學(xué)與學(xué)習(xí)國(guó)家重點(diǎn)實(shí)驗(yàn)室朱朝喆研究員czzhu@1精選2021版課件fMRI研究框架實(shí)驗(yàn)設(shè)計(jì) 被試招募與 掃描科學(xué)問題結(jié)果解釋

實(shí)驗(yàn)假設(shè) 數(shù)據(jù)統(tǒng)計(jì) 分析SPM,AFNI,FSL,VoxBo2精選2021版課件SPM版本歷史TheforthcomingversionisSPM5ThecurrentversionisSPM2PreviousversionsSPM2breleased21stNovember2002SPM99released25thJanuary2000SPM96released9thApril1997http://www.fil.ion.ucl.ac.uk/spm/3精選2021版課件線性代數(shù)統(tǒng)計(jì)理論GLM模型隨機(jī)場(chǎng)模型MR成像信號(hào)處理計(jì)算神經(jīng)解剖學(xué)神經(jīng)科學(xué)4精選2021版課件SPM數(shù)據(jù)分析基本流程5精選2021版課件預(yù)處理部分模型構(gòu)建與參數(shù)估計(jì)常用工具與參數(shù)設(shè)置6精選2021版課件講座提綱SPMI:PreprocessingSPMII:Single-subjectanalysesSPMIII:Groupanalyses7精選2021版課件SPMI:預(yù)處理1.2.3.4.Slicetiming–(獲取時(shí)間校正)Realignment-(頭動(dòng)校正)Normalisation-(空間標(biāo)準(zhǔn)化）Smoothing-(空間平滑)8精選2021版課件MRIvs.fMRI↑neuralactivity↑bloodoxygen↑fMRIsignalMRIfMRIoneimagehighresolution

(1mm)

lowresolution(~3mmbutcanbebetter)fMRI

BloodOxygenationLevelDependent(BOLD)signal indirectmeasureofneuralactivity

…manyimages(e.g.,every2secfor5mins)9精選2021版課件預(yù)處理–SliceTiming-SPM選擇參考slice拉齊其它slice10精選2021版課件預(yù)處理–Realign（頭動(dòng)校正）?不同scan之間像素對(duì)應(yīng)關(guān)系遭到破壞。?血液動(dòng)力學(xué)響應(yīng)被頭動(dòng)引起的信號(hào)淹沒。11精選2021版課件預(yù)處理–Realign（頭動(dòng)校正）剛體變換六個(gè)頭動(dòng)參數(shù)估計(jì):3個(gè)方向的平移(mm) 3個(gè)軸向的旋轉(zhuǎn)12精選2021版課件預(yù)處理–Realign-SPM將同一被試不同采樣時(shí)間點(diǎn)上的3D腦對(duì)齊13精選2021版課件空間標(biāo)準(zhǔn)化問題14精選2021版課件空間標(biāo)準(zhǔn)化問題個(gè)體大腦在形狀、大小等方面存在明顯差異，我們?nèi)绾芜M(jìn)行不同人之間的比較呢？…使不同被試腦圖像中的同一像素代表相同的解剖位置一個(gè)標(biāo)準(zhǔn)腦空間15精選2021版課件標(biāo)準(zhǔn)腦空間-Talairach坐標(biāo)系Source:BrainVoyagercourseslidesTalairach&Tournoux,1988?squishorstretchbraininto“shoebox”?extract3Dcoordinate(x,y,z)foreachactivationfocus16精選2021版課件使不同被試腦圖像中的同一像素代表相同的解剖位置粗配準(zhǔn)–仿射變換精配準(zhǔn)–非線性變換17精選2021版課件Why使不同被試腦圖像中的同一像素代表相同的解剖位置一個(gè)公共的標(biāo)準(zhǔn)空間How先使用簡(jiǎn)單的線性變換進(jìn)行粗配準(zhǔn)再用復(fù)雜的非線性變換精配準(zhǔn)Problems計(jì)算復(fù)雜度（高精度算法配準(zhǔn)一個(gè)腦需要幾個(gè)小時(shí)）個(gè)體之間的腦并非一一映射關(guān)系不可能有完全準(zhǔn)確的配準(zhǔn)Solutions對(duì)空間標(biāo)準(zhǔn)化后的腦圖像進(jìn)行適當(dāng)?shù)钠交褂米冃螆?chǎng)信息預(yù)處理–空間標(biāo)準(zhǔn)化–小結(jié)18精選2021版課件預(yù)處理–空間標(biāo)準(zhǔn)化-SPM使不同被試腦圖像中的同一像素代表相同的解剖位置將每個(gè)個(gè)體腦放入一個(gè)公共的標(biāo)準(zhǔn)空間19精選2021版課件TemplateNormalisedImage預(yù)處理–空間標(biāo)準(zhǔn)化–結(jié)果20精選2021版課件

空間平滑的問題使殘差項(xiàng)更符合高斯分布假設(shè)減少標(biāo)準(zhǔn)化后剩余的個(gè)體間差異提高信噪比5-5021精選2021版課件預(yù)處理–空間平滑-SPM22精選2021版課件SPM預(yù)處理部分小結(jié)1.2.3.4.Slicetiming–(adjusttimedifferenceamongdifferentslice)Realignment-(adjustformovementbetweenslices)Normalisation-(warpfunctionaldataintotemplatespace)Smoothing-(toincreasesignaltonoiseratio)23精選2021版課件LectureOutlineSPMI:PreprocessingSPMII:Single-subjectanalysesSPMIII:Groupanalyses24精選2021版課件Single-subjectAnalyses基本過程與原理GLMPrincipleofGLMDesignMatrixSolutiontoGLMEffectofInterestandstatistics25精選2021版課件個(gè)體水平分析的基本過程與目的實(shí)驗(yàn)設(shè)計(jì)個(gè)體掃描個(gè)體激活區(qū)檢測(cè)

SpatialMemoryCondition

500msec 200msec 3000msec Time 1500msec 500msec 3000msec 200msecSpatialControlCondition

1500msec

對(duì)這個(gè)被試，你感興趣的effect在那些腦區(qū)出現(xiàn)，其強(qiáng)度如何？26精選2021版課件Single-subjectAnalyses基本過程與原理GLMPrincipleofGLMDesignMatrixSolutiontoGLMEffectofInterest&StatisticsExample27精選2021版課件Single-subjectAnalyses基本過程與原理GLMPrincipleofGLMDesignMatrixSolutiontoGLMEffectofInterest&StatisticsExample28精選2021版課件InMatrixFormGLM的數(shù)學(xué)表示………β1:βl::…:…:::YJxJ1…xJl…xJLβLεJX

Y=觀測(cè)數(shù)據(jù)

×設(shè)計(jì)矩陣

β參數(shù)+ε

殘差x1lx1L

ε1恐懼Y1

:x11:Yj=xj1×β1+...+xjl×βl+...+xjL×βL+εj

::Y1x11…x1l

:YJ:x…:Yj=xj1…xjlJ1x1L

:xJlxjL

ε1

:xJL+εjεJYY^29精選2021版課件Single-subjectAnalyses基本過程與原理GLMPrincipleofGLMDesignMatrixSolutiontoGLMEffectofInterest&StatisticsExample30精選2021版課件TimeTimeGLM：設(shè)計(jì)矩陣XX2X1

Y=X?β+εSPMrepresentstimeasgoingdownSPMrepresentspredictorswithinthedesignmatrixasgrayscaleplots(whereblack=low,white=high)overtimeSPMincludesaconstanttotakecareoftheaverageactivationlevelthroughouteachrunXIntensityY31精選2021版課件G(刺激因素）DesignmatrixXG1H(干擾因素)H1Globalactivity:E.g.headmotionparametersHcLineartrendsGcstimulusGLM：設(shè)計(jì)矩陣X的結(jié)構(gòu)32精選2021版課件血氧系統(tǒng)對(duì)單次刺激的響應(yīng)33精選2021版課件刺激序列HRF設(shè)計(jì)矩陣中的刺激因素X34精選2021版課件G(stimulating)DesignmatrixXG1H(non-interesting)H1E.g.(1)headmotionparameters (2)breathing (3)heartbeatHcLineartrends

duetoMRIscannerGlobalactivity:GcstimulusGLM：設(shè)計(jì)矩陣X的結(jié)構(gòu)35精選2021版課件為什么要考慮這些干擾因素？LinearTrendProbableRespirationArtifactheadmotion parametersEffect/Error36精選2021版課件Single-subjectAnalyses基本過程與原理GLMPrincipleofGLMDesignMatrixSolutiontoGLMEffectofInterestandstatistics37精選2021版課件0100-10+10012-0.01+0.01=+*5+…+Y=X1*β1+…+Xn*μ+e*50FittingXtoYgivesyouoneβ(parameterestimate)foreachcolumnofX,aμande.BetasprovideinformationaboutfitofregressorXtodata,Y,ineachvoxel38精選2021版課件GLM求解的幾何表示：勾股定理E用X線性組合Y^近似表達(dá)Y39精選2021版課件Single-subjectAnalyses基本過程與原理GLMPrincipleofGLMDesignMatrixSolutiontoGLMEffectofInterest&statistics多重比較Example40精選2021版課件

構(gòu)造Contrast對(duì)感興趣的解釋變量進(jìn)行比較X2X1Y=X?β+ε

=β1?X1+β2?X2+β3?X3+…+βN?XN+ε

[X1X2X3…XN] [β1β2β3…βN]

T檢驗(yàn)：構(gòu)造Contrast向量

F檢驗(yàn)：構(gòu)造Contrast矩陣實(shí)驗(yàn)設(shè)計(jì)=〉感興趣effect=〉contrast所以contrast在數(shù)據(jù)采集之前就定下了！41精選2021版課件

本質(zhì)Effects?解釋空間Xs?contrast向量[1-1]x1–x2??

Ex1–x2

(x1–x2)42精選2021版課件Single-subjectAnalyses基本過程與原理GLMPrincipleofGLMDesignMatrixSolutiontoGLMEffectofInterest&statisticsMultipleComparisons43精選2021版課件TimeY=X?β+εIntensityYPreprocessing...TheProblemofMultipleComparisons

T>ToPo=0.0144精選2021版課件200activated噪聲腦的“激活”

P=0.0120,000voxs

噪聲腦45精選2021版課件怎么辦？200activated2activated200activated 5activated46精選2021版課件200activated200activatedUncorrectedp=0.01

我在進(jìn)行探 索性研究！探索性研究47精選2021版課件Bonferronicorrection 最嚴(yán)格的校正200activated2activatedonevoxelTypeIerrorp=？

numberofvoxels：N=50,000 overallcorrectdetection=(1-p)(1-p)…(1-p)=(1-p)N

overallTypeIerror=1-(1-p)N=~Np DesiredoverallTypeIerror：Np=.05 RequiredonevoxelTypeIerrorp=.05/50,000=.00000148精選2021版課件BonferroniCorrection的思想及其在fMRI數(shù)據(jù)分析中的問題Bonferroni校正的假設(shè)pvoxel=poverall/NN為獨(dú)立觀測(cè)個(gè)數(shù)相鄰體元的BOLD信號(hào)會(huì)相互獨(dú)立的嗎？頭動(dòng)等噪聲對(duì)同一腦區(qū)的影響很相似BOLD信號(hào)本身就對(duì)應(yīng)著一定空間范圍預(yù)處理中的平滑49精選2021版課件SPM中的多重比較校正的原理根據(jù)數(shù)據(jù)的空間相關(guān)程度計(jì)算獨(dú)立觀測(cè)個(gè)數(shù)（獨(dú)立比較的次數(shù)Nindepentent）根據(jù)整體虛警概率poverall和Nindepentent得到單個(gè)體元的pvoxel值pvoxel=poverall/Nindepentent50精選2021版課件SPM個(gè)體激活區(qū)檢測(cè)基本過程51精選2021版課件個(gè)體水平effect計(jì)算的SPM實(shí)現(xiàn)

（個(gè)體激活區(qū)檢測(cè)）模型定義DesignMatrixSpecification

數(shù)據(jù)定義參數(shù)估計(jì)

DataSpecificationParameterEstimation統(tǒng)計(jì)結(jié)果Result52精選2021版課件參數(shù)估計(jì)常用工具與 參數(shù)設(shè)置

預(yù)處理部分

First-level模型構(gòu)建與Second-level53精選2021版課件LectureOutlineSPMI:Intro,PreprocessingSPMII:Single-subjectanalysesSPMIII:Groupanalyses54精選2021版課件Howdowecompareacrosssubjects?建立不同人之間的可比性NormalizationROI多個(gè)被試的統(tǒng)計(jì)分析Fixed-effectsModelRandom-effectsModel55精選2021版課件Fixed-effectsModelAssumethattheexperimentalmanipulationhassameeffectineachsubjectUsesdatafromallsubjectstoconstructstatisticaltestAveraging/connectingacrosssubjectsbeforeat-testSensitivetoextremeresultsfromindividualsubjectstrongeffectinonesubjectcanleadtosignificanceevenwhenothersshowweakornoeffectsAllowsinferencetosubjectsampleyoucansaythateffectwassignificantinyourgroupofsubjectsbutcannotgeneralizetoothersubjectsthatyoudidn’ttestHowaboutthepopulation?56精選2021版課件RandomeffectanalysisAssumesthateffectvariesacrossthepopulationAccountsforinter-subjectvarianceinanalysesAllowsinferencestopopulationfromwhichsubjectsaredrawnEspeciallyimportantforgroupcomparisonsRequiredbymanyreviewers/journals57精選2021版課件SPM雙層統(tǒng)計(jì)First-level：個(gè)體水平effect計(jì)算Second-level：群體水平effect計(jì)算58精選2021版課件SPM個(gè)體激活區(qū)檢測(cè)基本過程59精