 |
 |
I’m looking for suggestions on how to analyze fMRI responses in a slow event-related design where we predict (based on previous studies and ICA + ROI timecourse ploting of our data) that different sets of regions will be active early vs. late in the trial.
The task involves presenting a phrase to cue retrieval of a specific autobiographical memory (positive, negative, or neutral), followed 16 seconds later by 12 seconds of rest/intertrial interval. There’s good reason to predict that early responses reflect memory access (hippocampal activation, etc.) whereas the later responses in the trial reflect maintenance and elaboration of the memory, recruiting other regions.
I haven’t been able to find much literature specifically discussing the analysis and the pros and cons of designs where there is no jittering between successive long events (8 seconds or more), where the order of the two events is necessarily fixed, as in this case. Another issue is that the component events of the trial will surely vary in onset and duration across stimuli, conditions , and subjects, so a Procrustean approach of assuming two component boxcar-like regressors of equal duration is surely an oversimplification. Yet, a previous study used this approach (with adjustment of the regressor length based on subject responses indicating the onset of retrieval), and found very sensible and interpretable results, that meshed with an ROI-based timecourse analysis.
A related approach that’s been suggested is to model the 16-second retrieval period with a “sustained” regressor (16-second boxcar) to assess retrieval processes that persist across that period, and a second regressor modeling intra-trial linear increase/decrease. This would finesse an arbitrary division of the trial, and is probably closest to the hypothesized processes. However, the pre-fab SPM function that used to do sustained/transient block responses isn’t in SPM8, and it isn’t completely clear how to deal with setting it up, orthogonalization of the regressors, etc., or whether it's the most appropriate approach.
I’d be happy to hear other suggestions or references.
--Stephan