Experimental design - Brain Research Imaging Centre Edinburgh

Experimental designAlexa MorcomEdinburgh SPM course 2013Thanks to Rik Henson, Thomas Wolbers, Jody Culham, andthe SPM authors for slides

Overview of SPMImage Image timeseriestimeseriesDesign Design matrix matrixContrastsPreprocessingGenerallinearmodelSPMsTemplateKernel KernelVariancecomponentsThresholding

Overview• Categorical designs• Factorial designs• Parametric designs• fMRI adaptation• Control condition• Paradigm timing

Isolating a processSubtraction logic and assumption of pure insertion• Compare task conditions differing in a single process• Measure the time the process takes• Assume that addition of the component processdoes not alter other task componentsDonders (1898-9)

Pure insertion

Cognitive subtraction

Cognitive subtraction

Categorical designsPrinciple• Subtract two conditions to isolate a process• Assume that addition of the component processdoes not alter other task components• So adding the process into different tasks shouldproduce the same change in activity• (and: meaningful cognitive theory)

Categorical designsSimple subtraction• Detect regions specialised for a function bytesting for activation difference

Categorical designsSerial subtraction• Several cognitive processes in picture namingSee Friston et al. chapter in Human Brain Function (I)

Experimental designCategorical designs

Problem:Categorical designs

Factorial designsInteractions• The whole (task) is more than the sum of its(interdependent) processes• A modulates B• Vary A and B independently

Factorial designs

Conjunction designTwo task pairs• B – viewing concrete objects and saying “yes”• C – naming concrete objectsDifference = phonological retrieval PLUS interactionwith object recognition• B2 – viewing coloured shapes saying “yes”• C2 – naming colour of coloured shapesDiff = phonological retrieval PLUS interaction withvisual analysis

Conjunction designOverlap isolates the process of interest• Phonological retrieval• NOT its interactions with visual processingOverlap of 4 subtractionsPrice & Friston (1997)

Conjunction designsDetecting overlapping processing• Encoding faces, different objects• Reactivation of same regions when face,object memories retrievedMVPA recall studyPolyn et al. (2005)

Parametric designsA continuously varying parameter• Systematic variation in activity with processengaged to varying degrees• Specific: e.g. Linear? Quadratic?• Avoids pure insertion but does assume noqualitative change in processing• Often less sensitive

Parametric designsARest + 5 rates of auditory word presentationBPET study• Auditory words,varying rate• Linear relationshipof rate withactivity in primaryauditory cortexPrice et al. (1992)

Parametric designsModel based fMRI• Computational model provides neurometricfunction e.g. Rescorla-Wagner predictionerror• Model comparisonGlascher & O’Doherty (2010)

Parametric factorial designsPsychophysiological interaction in V5z = -9Friston et al. (1997)• V1 activity =parametric(physiological)predictor• Attention tomotion =categorical(psychological)predictor

fMRI adaptation• Repetition suppression• = reduced BOLD response to repeated stimuli• Accompanies priming (behavioural)

fMRI adaptationRepetition suppression as a tool• fMRI – typical voxel = 10,000s of neurons• FFA – a mix, tuned to diff. face orientations?• Or: all viewpoint-invariant?

fMRI adaptationRelease from adaptation => sensitivity to the changed feature

fMRI adaptationOrientation tuning in human LOCIdent. Trans. Rotate

Control conditionProblem• fMRI is a contrastive method – for manydesigns, you need a control• ‘Rest’ isn’t no processing in many areas

Control conditionDifferent stimuli and taskDifferent stimuli similar task+‘Marilyn’Wonder if I leftthe gas on…?‘Female’‘Female’Same stimuli different taskSimilar stimuli same task‘Female’‘Seen before’‘Female’‘Female’ Choice of a baseline depends on your question!

Control conditionStark & Squire (2001) PNAS

Paradigm timingAnalysis of whole blockLarge effects (=efficient)Optimal length = 16 sec(sluggish BOLD vs. lowfrequency confounds)Analysis of single itemsSmaller effectsSOA from min ~= 2 sec

Paradigm timingAdvantages of event-related design• Intermixing of conditions avoids unwantedpsychological effects e.g. habituation, expectancy,loss of concentration

BlockedDataModelO = Old WordsN = New WordsO1O2O3RandomisedN1N2N3O1 N1 O2 O3N2

Paradigm timingAdvantages of event-related design• Intermixing of conditions avoids unwantedpsychological effects e.g. habituation, expectancy,loss of concentration• Post-hoc classification of trials, e.g. Subsequentmemory effect

Paradigm timing

Paradigm timingAdvantages of event-related design• Intermixing of conditions avoids unwantedpsychological effects e.g. habituation, expectancy,loss of concentration• Post-hoc classification of trials, e.g. Subsequentmemory effect• Some events can only be indicated by subject atparticular time e.g. Spontaneous perceptual changes

Number of Perceptual Reversals201510500 5 10 15 20Inter−Reversal Time (s)25201510500 5 10 15 20Inter−Reversal Time (s)

Paradigm timingAdvantages of event-related design• Intermixing of conditions avoids unwantedpsychological effects e.g. habituation, expectancy,loss of concentration• Post-hoc classification of trials, e.g. Subsequentmemory effect• Some events can only be indicated by subject atparticular time e.g. Spontaneous perceptual changes• Some events cannot be blocked, e.g. oddball

SummaryA few principles, one main take-home message• Different designs for different questionsWant to know more?• Temporal design efficiency• Design optimisation Advanced course (Wednesday)