1 – – 1 AM Regression - 1 AM Regression - 1 mplitude M odulated (or M odulation) • AM = A Have some extra data measured about each response to a stimulus, ★ and maybe the BOLD response amplitude is modulated by this ★ Reaction time; Galvanic skin response; Pain level perception; Emotional valence ( happy or sad or angry face?) • Want to see if some brain activations vary proportionally to ( ehaviorial I nformation) uxiliary ABI this A B Discrete levels (2 or maybe 3) of ABI: • ★ Separate the stimuli into sub-classes that are determined by the ABI on off ” , maybe?) ” “ ( “ and Use a GLT to test if there is a difference between the FMRI responses in ★ the sub-classes \ ... 3dDeconvolve 1 'On' \ -stim_times 1 regressor_on .1D 'BLOCK(2,1)' -stim_label .1D 'BLOCK(2,1)' -stim_label 2 'Off' \ -stim_times 2 regressor_off -glt_label -gltsym 'SYM: +On | +Off' 1 'On+Off' \ 2 'On-Off' 'SYM: +On -Off' -gltsym ... -glt_label off tests for any activation in the “ on or “ either ” conditions ” On+Off “ ” between “ on ” and “ off ” conditions “ On-Off ” tests for differences in activation to threshold on both statistics at once to find a conjunction 3dcalc Can use

2 – 2 – AM Regression - 2 Continuous levels (or several finely graded) • ABI Want to find active voxels whose activation level also depends on ABI ★ ★ 3dDeconvolve is a linear program, so must make the assumption that the change in FMRI signal as ABI changes is linearly proportional to the changes in the ABI values • Need to make 2 separate regressors (the usual ★ analysis ) One to find the mean FMRI response -stim_times One to find the variations in the FMRI response as the ABI data varies ★ The second regressor should have the form • K ! a ( # ) " ! t ( h = ) t ( r ) a $ k 2 A k M k = 1 th value = value of k Where ABI value, and a ★ a is the average ABI k ) for first regressor • is standard activation map Response ( β • Statistics and β for second regressor make activation map of whose places BOLD response changes with changes in ABI are Using 2 regressors allows separation of voxels that are active but ★ ABI modulated by the detectably are not ABI-sensitive from voxels that

3 – 3 – AM Regression - 3 New feature of -stim_times_AM2 3dDeconvolve • : is very similar to standard -stim_times Use • -stim_times_AM2 1 times_ABI .1D ' BLOCK(2,1) ★ ' The .1D file has time entries that are “ married ” ★ times_ABI 27*2 39*5 23*4 10*5 to ABI values: 17*2 32*5 * 24*3 37*5 41*4 16*2 files ★ Such files can be created from 2 standard ASCII .1D using the new 1dMarry program -divorce option can be used to split them up o The 3dDeconvolve automatically creates the two regressors • ( unmodulated and amplitude modulated) option to get statistics for activation of the pair of ★ Use -fout (i.e., testing null hypothesis that β regressors weights are zero: both ABI-proportional signal change) or that there is no ABI-independent option to test each β weight separately ★ Use -tout regressor matrix columns to see each X 1dplot ★ Can

4 – 4 – AM Regression - 4 new The AM needs some practical user • , and so feature is standard practice “ experiences before it can ” be considered ★ In particular: don ʼ t know how much data or how many events are statistics needed to get good ABI-dependent remove potential confounds in You could also use AM regression to ★ FMRI responses that co-vary with some external parameter • If you want, -stim_times_AM1 is also available regressor ★ It only builds the proportional to ABI data directly, with no K removed: mean " ! t ( a = # ) t ( r ) h $ k M A k 1 k 1 = t imagine what value this option has, but you never know ★ ... ʼ (if you Can can think of a good use, let me know) • Future directions: ( insert ★ Allow more than one amplitude to be married to each stimulus time polygamy/polyandry joke here) obligatory ABI types at once is o too many? I don ʼ t know. How many How to deal with unknown nonlinearities in the BOLD response to ABI ★ I don ʼ t know. (Regress each event separately, then compute MI?) values? ★ Deconvolution with amplitude modulation? Requires more thought.

5 – 5 – AM Regression - 5 10*1 30*2 50*3 70*1 90*2 110*3 130*2 150*1 170*2 190*3 210*2 230*1 = Timing: AM.1D 3dDeconvolve -nodata 300 1.0 -num_stimts 1 \ • -stim_times_AM1 1 AM.1D 'BLOCK(10,1)' -x1D AM1.x1D 1dplot AM1.x1D'[2]' • model of signal AM1 (modulation = ABI) 3dDeconvolve -nodata 300 1.0 \ • -num_stimts 1 \ -stim_times_AM2 1 \ AM.1D 'BLOCK(10,1)' \ -x1D AM2.x1D 1dplot -sepscl \ • AM2.x1D'[2,3]' AM2 model of signal: is 2D sub-space spanned by these 2 time series

6 – – 6 AM Regression - 6 (NIDCD; PI=Al Braun) First actual user: Whitney Anne Postman • Picture naming task in aphasic stroke patient • • ABI data = number of alternative names for each image (e.g., ” & “ porch ” & “ veranda ” , vs. “ strawberry ” ) , from 1 to 18 balcony “ 8 imaging runs, 144 stimulus events • • 2 slices showing activation map for BOLD responses proportional to ABI ( β ) AM2 • What does this mean about the brain and aphasia? Don ʼ t ask me!

7 – – 7 AM Regression - Updates: Dec 2008 -stim_times_AM* options (and program ) • The 1dMarry have been modified to allow the input of multiple amplitudes with each time For example: • 37.2*1,2,3 42.6*-1,7,4 53.7*2,-6,1 • • 3 extra amplitudes per time point -stim_times_AM2 will generate 4 regressors • 1 regressor for the constant magnitude FMRI response • • 1 regressor for each of the extra amplitudes -stim_times_AM1 • still builds only 1 regressor, as before BLOCK (say) is modulated by sum of all • Amplitude of each extra amplitudes provided You can use -stim_times_AMx as a synonym for - • stim_times_AM2 , if you prefer

8 – – 8 Regression - DM Dec 2008 a new variant of the For use with BLOCK • -stim_times_AM* response model function is available: dmBLOCK duration modulation instead of (or on top of) amplitude modulation ” • “ • Example: 10*1:5 30*2:10 50*3:15 70*1:5 File A1.1D • = 2 extra parameters per time: amplitude modulation:duration • Last parameter always duration • : Use as the separator • ʻ ʼ • Previous = amplitude modulation 3dDeconvolve \ • -nodata 100 1.0 \ -num_stimts 1 \ -polort -1 \ -stim_times_AM1 1 A1.1D \ dmBLOCK -x1D A1dm.x1D • 1dplot A1dm.x1D

9 – – 9 Regression - DM 2 model Same thing, but with • AM2 3dDeconvolve \ • -nodata 100 1.0 \ -num_stimts 1 \ -polort -1 \ -stim_times_AM2 1 A1.1D \ dmBLOCK -x1D A2dm.x1D 1dplot A2dm.x1D • ʼ t require an amplitude modulation • dmBLOCK doesn parameter, but will use one (or more) if present last • Duration parameter is always the parameter married ʼ character : ʻ to the time, and is separated by a Future dream: may have more nonlinearly modulated • response model functions — any ideas?