Functional Magnetic Resonance Imaging (fMRI) - PowerPoint Presentation

1. Functional Magnetic Resonance Imaging (fMRI)Jan Petr

2. MRI quick summarySpin property of hydrogen atomsUsing strong B0 magnetic field1.5 T, 3T – clinical scanners7T – experimental scanners

3. MRI quick summaryImaging magnetic properties of tissueProton densityT1-weighted relaxationT2-weighted relaxationT1PD T2

4. Brain imaging with different modalitiesStructureSoft tissueBonesVesselsPhysiologyMetabolismFunctionPET (Positron emission tomography)CT (Computed tomography)MRI

5. Functional MRIImage brain activitySpatial resolution ~mmTemporal resolution ~s

6. Brain regionsAnatomical regionsIndividual differencesize?shape?topology?Functional regions

7. Brain regionsExamples of brain activation regionsSensoryMotorLanguageVisionTouchFinger tappingPicture namingListening to wordsReversing checkerboardPassiveactiveactive passivepassive

8. Brain anatomyNeurons and glial cellsNeurons communicate through axons Through electrochemical processes

9. Brain anatomyGray matterConsists mostly of neuronsWhite matterConsists mostly of axonsGray matterWhite matter

10. Neuronal activationIntegrative and signalling activityChange cell membrane potentialRelease of neurotrasmittersIonic pumps to restore concentration gradientsRequires glucose and oxygen

11. Brain vasculatureBlood supplies brain with oxygen and glucoseInternal carotid and vertebral arteriesFurther branching to microvessels and capillaries

12. Neurovascular couplingNeurovascular couplingVasoactive substances  Dilate vessels Reduces resistance Increase blood flow

13. fMRI physiologyWhat is measured in fMRI?Electrical impulses?Neurotransmitters?Blood perfusion?Blood perfusion through the level of oxygenation

14. History of BOLD imagingBOLD – Blood Oxygenation Level DependentOgawa et al., 1990Mice and rats at 7T MRIContrast on gradient-echo images influenced by proportion of oxygen in breathing gasIncreasing oxygen content  increased contrastOgawa et al., 1992Humans at 4T MRIVisual stimulationChanges of contrast in visual cortex

15. BOLD signal and T2*T2* relaxation – decay of signal after excitationTwo components of T2* :Intermolecular interactions  dephasing  T2 signal decayMacroscopic magnetic field inhomogeneity  dephasing  T2’ decay. 

16. BOLD signal and T2*Why does blood oxygenation affect the BOLD MRI signal?Hemoglobin contains iron to bind the oxygenOxyhemoglobin (oxHb) is diamagnetic Deoxyhemoglobin (dxHb) is paramagneticHigher dxHb concentration  increased magnetic susceptibility  increased magnetic field inhomogeneities  decrease T2*  lower BOLD MRI signal

17. Hemodynamic responseNeuronal activity  Increased O2 metabolism  Increased dxHb  lower BOLD signal? Neurovascular coupling  Vessel dilation  increased CBF dxHb concentration decreases  higher BOLD signal

18. Hemodynamic responseNeuronal activityGlucose and oxygen metabolismDecay Time (T2*)T2* weighted image intensityMagnetic field uniformityCerebral blood volume (CBV)Blood oxygenationCerebral blood flow (CBF)Brain functionMetabolic ratesPhysiological effectsPhysical effectsMR properties-+

19. Hemodynamic responseDelay in BOLD signal change after activationInitial dip – increase in oxygen consumption before CBF increaseUndershoot – CBF decrease faster than CBVPeakPeakRiseRiseBaselineBaselineSustainedresponseUndershootUndershootInitial dip

20. fMRI experimental designGoal: To detect what regions/voxels are active during a specific task

21. What sequence should be used for fMRINeuronal response - 200-500msHemodynamic response – ~s Standard whole brain sequence~1mm spatial resolutionTime resolution ~minsFast single shot sequencesEcho planar imaging (EPI)500ms-2s acquisition

22. fMRI task designCreate a desired cognitive stateDetect brain signals associated with that state

23. Types of fMRI designsBlock-designDetection powerEvent-related designMore flexibleMixed designBlock designEvent-related design

24. Readout in fMRI design↑ spatial resolution:↓ time resolution↓ coverage (number of slices)↑ temporal resolution requires:↓ spatial resolution↓ coverage (number of slices) ↑ SNR (signal-to-noise ratio):↓ Decreased spatial resolution↑ Increased scan time via averagingSpatial resolutionTemporal resolutionSNR

25. fMRI study designBOLD signal – combination CBV, CBF, CMRO2Observe change of BOLD signal as a reaction on a task or eventStimulusHRFExpectedresponse

26. I have my data, now what?Data pre-processingStructural MRI functional MRI

27. Why pre-process fMRI dataData are noisy (task-related change <5%)Subjects moveThings change during the experimentPreprocessing: Increase signal to noise ratio Helps to meet assumptions for statistical analysis

28. Subject motionCorrect for head motion6 parameters rigid transformation3 rotations 3 translationsLie very stillExclude subjects

29. Spatial normalizationRegister functional vs. anatomical per subjectRegister to average brain (MNI)Larger populationHigher powerWithin-subjectBetween-subject6 DOF> 6 DOF

30. Temporal filteringTemporal drift from scannerHigh-pass filterPhysiological cycles (cardiac, respiratory)

31. Spatial filteringConvolution with a Gaussian kernelImprovesSNRSpecificityReducesSpatial resolutionSensitivity

32. Is there an activation?A finger tapping example

33. A simple fMRI experimentPassive tapping vs rest (7 cycles)Blocks of 6 scans per cycleIs there a change in the BOLD response between finger tapping and rest?Stimulus function

34. A simple fMRI experimentActivation  compare:Magnitude of responseMeasurement noiseT-testStimulus functionSignal from one voxelCompare tap in green vs rest

35. General linear modelExperimental data (Y) - linear combination (β) of different model factors (x), along with uncorrelated noise (ε)Testing slope (β) against null hypothesisY = βx + β0 + εβ0

36. General linear model for fMRI Y = X * β + εtimepointsObserved data (known) BOLD signal in a single voxelDesign matrix (known)Components that can explain the dataModel parameters(unknown)Contribution of each component of X to YErrorDifference between the observed data and model prediction

37. GLM example: DesignBlock design, language taskWord generation (noun presented, verb generated)Word shadowing (verb presented, thinking on it)RestDesign matrix:generationshadowingrest

38. GLM example: Estimating betasFitting model to data – ordinary least squares – minimizing  ≈β1∙+ β2∙+ β3∙2340101012

39. ≈β1∙+ β2∙+ β3∙2340101012003GLM example: Estimating betasSuboptimal fit 

40. ≈β1∙+ β2∙+ β3∙23401010120.830.162.98GLM example: Estimating betasActive in word generation 

41. ≈β1∙+ β2∙+ β3∙12301010120.680.822.17GLM example: Estimating betasActive in word generation and shadowing 𝛽=[0.68, 0.82, 2.17]

42. ≈β1∙+ β2∙+ β3∙12301010120.030.062.04GLM example: Estimating betasVoxel not active 𝛽=[0.03, 0.06, 2.04]

43. GLM example: Voxelwise fit Calculate fit for every voxelbeta1beta2beta3......Time-seriesresiduals  

44. GLM example: Significance Which of these series should we trust?Noise, effect size, number of measurementsβ1=1σ=0.2n=60β1=1σ=0.5n=60β1=0.3σ=0.2n=60β1=1σ=0.2n=15

45. GLM example: Contrast Weights c of model parameters βc = [c1 c2 c3] for β = [β1 β2 β3]c = [1 0 0]Active in word generatingc = [1 -1 0]More active in generating than in shadowing

46. GLM example: Hypothesis testing Null hypothesis (H0) – there is no effectAlternative hypothesis (Ha) – we find the effect in dataReject the null hypothesis  activation 

47. GLM example: t-contrastfollows Student’s distribution (N-1 degrees of freedom)Probability that the null hypothesis is truep-value <0.05 we reject the null hypothesis 

48. GLM example: t-contrast example

49. GLM example: t-contrast example Voxels active in word generationc=[1 0 0]

50. GLM example: t-contrast example Voxel active more in generating than shadowingc=[1 -1 0]

51. fMRI applicationsSurgery planningVolunteerPatient with glioblastoma

52. fMRI applicationAddictionUnderstanding of brain effects of long-term useDevelopment of treatment strategies for abusersPharmacological studiesEffects on cognitionNeuropsychological disordersDisease markers may help in treatmentAging and brain developmentNormal and pathological changes

53. fMRI summarySimple and non-invasiveVery good time and spatial resolutionWide range of applicationsProblems with noiseLimited clinical use