FMRI Experimental DesignLila DavachiDepartment of Brain & Cognitive Sciences, M.I.T.Because fMRI BOLD data is not anBecause fMRI BOLD data is not anabsolute measure of neuronal activity,absolute measure of neuronal activity,all study designs must provide theall study designs must provide theopportunity to statistically contrast theopportunity to statistically contrast theneuronal activity of interest with aneuronal activity of interest with asuitable rest or background conditionsuitable rest or background condition..Thus, study design is Thus, study design is of paramount importance.of paramount importance.Why?Why?HypothesisHypothesisHow?How?fMRI Study DesignfMRI Study DesignWhere?Where?NeuroanatomyNeuroanatomyWhat?What?BehaviorBehaviorThanks to Chantal Stern43 * 7 = ?Key Points• What can fMRI tell you?• Always comparing across conditions• Characteristics of the hemodynamic response(HRF) and how this affected the sequentialdevelopment of fMRI paradigms and influencesstudy design• Sense of important design issuesWhat (good) is fMRI?What it can tell you:• Relative local “neural” activity (LFP’s ?)• NOT absolute neural activity• NOT excitation vs inhibition• NOT about necessity of a given region for a task• NOT fine-grained temporal informationKey Points• What can fMRI tell you?• Always comparing across conditions• Characteristics of the hemodynamic response(HRF) and how this affected the sequentialdevelopment of fMRI paradigms and influencesstudy design• Sense of important design issuesSubtraction ParadigmDonder’s method:Ex: How to measure time of a mental transformation?A random series of A’s and B’s presented and the subject must:1. Respond whenever an event occurs (RTi)2. Respond only to A not to B (RTii)3. Respond X to A and Y to B (RTiii)RTi = RT(detect) + RT(response)RTii = RT(detect) + RT(discrimination) + RT(response)RTiii = RT(detect) + RT(discrimination) + RT(choice) + RT(response)THUS, RT(discrimination) = RTii - RTi RT(choice) = RTiii - RTiiCriticisms of Subtraction Paradigm1. That we already know what ‘counts’ as a single mentalprocess (i.e. choice is a single mental process?)2. Assume that adding components does not affect otherprocesses (i.e. assumption of pure insertion)THUS, one should pick tasks that differ along ONEdimension (either change the task OR the stimuli but notBOTH!)And a resting baseline is good to include, however, theinterpretation should be taken lightly…(more later)The loose task comparisonThe loose task comparisonDoes Does notnot hold all variables constant BUT: hold all variables constant BUT:(1) Uses a low level reference task(1) Uses a low level reference task(2) Allows the data to be examined for predictable(2) Allows the data to be examined for predictablestimulus or response drivenstimulus or response driven activations activations(3) Allows the more extensive activation pattern to(3) Allows the more extensive activation pattern tobe observedbe observedThe “loose” TaskThe “loose” TaskComparisonComparisonSTISTISTIPREPREPREDOGDOGDOG++++++++++++GREGREGRESTISTISTIPREPREPREPAINTPAINTPAINTSTRSTRSTRSTISTISTIPREPREPREEGGEGGEGGTASK 1TASK 1TASK 2TASK 2The tight task comparisonThe tight task comparisonTry to hold all variables constant including:Try to hold all variables constant including:•• Stimulus display (nominally or statistically) Stimulus display (nominally or statistically)•• Response and response selection characteristics Response and response selection characteristics•• Performance level- especially if comparing cohorts Performance level- especially if comparing cohorts•• Eye movements Eye movements•• Emotional state (minimize anxiety and boredom) Emotional state (minimize anxiety and boredom)The “tight” TaskThe “tight” TaskComparisonComparisonSTISTISTIPREPREPREDOGDOGDOG++++++++++++GREGREGRESTISTISTIPREPREPREPAINTPAINTPAINTSTRSTRSTRSTISTISTIPREPREPREEGGEGGEGGTASK 1TASK 1TASK 3TASK 3TASK 2TASK 222 mmiinnuuss 11TASK 1TASK 1TASK 2TASK 2BRAIN AREAS THAT DIFFERBRAIN AREAS THAT DIFFERALL ACTIVE BRAINALL ACTIVE BRAINAREASAREASThanks to Randy BucknerExample...Interested in semantic processing and how it affectsmemory...Parameters to specify in any experiment1. Subjects: normal vs special populations2. What part of brain look at? How many slices canyou have for your TR?3. Choosing your TR: How often can you take a full setof pictures4. What coil will you use?surface coils: higher SNR, only partial coveragehead coils: lower SNR, complete coverage5. Toggle many times between conditions within a scan6. Run as many scans as possible within a subjectKey Points• What can fMRI tell you?• Always comparing across conditions• Characteristics of the hemodynamic response(HRF) and how this affected the sequentialdevelopment of fMRI paradigms and influencesstudy design• Sense of important design issuesVisual Stimulation - 2 sec Flashes0 10 20 30 40 50 60 70 80 90 100 110 120TIME (seconds)Percent Signal Change0.01.02.03.04.0-1.0[Blamire, Ogawa et al., PNAS, 1992]Visual Cortex During Brief Visual Stimulation0.01.02.05 0 5 10 15341001000msecTIME (seconds)msecmsec[Courtesy of Robert Savoy, Kathleen O’Craven MGH-NMR Center]Blocked design fMRIBLOCKED:HORSE(abstract or concrete?)love(upper or lowercase?)“Blocked” fMRI:Memory ParadigmSTIPREHORSE+ +0 8 74 118 140 184 206 250GRESTIPRELOVESTRSTIPREHUMORTIME (SEC)52++STRSTIPRECHAIR+[Wagner et al., OHBM, 1998]Typical Blocked-Design Response-0.2500.250.50Percent Signal Change0 20 40 60 80 100 120 140 160 180 200 220 240TimeHORSE LOVE HUMOR CHAIR13 SlicesPerBrain ImageThanks to Robert Savoy80 brain imagesper 4 minuterunThanks to Robert SavoyFor purposes of illustration……. Thanks to Robert SavoyExamine thedata fromone sliceof the brainas a functionof timeThanks to Robert SavoyThanks to Robert SavoyAre these voxellevels statisticallydifferent from each other?CombinetheseCombinetheseThanks to Robert SavoyCombinetheseCombinetheseConsider EACH voxelacross all time pointsThanks to Robert SavoyTypical Blocked-Design Response-0.2500.250.50Percent Signal Change0 20 40 60 80 100 120 140 160 180 200 220 240TimeHORSE LOVE HUMOR CHAIREvent-Related fMRIBLOCKED:SPACED EVENT-RELATED:16 sec“Spaced Event-Related” fMRI:Language Paradigm[Buckner, Bandettini et al., PNAS, 1996]TIME (seconds)0 32 64 96COUPRETRAAFTSTAPELDRI“Single-Trial” Response Across a Run“Single-Trial” Response Across a RunTIME (SEC)TIME