Trial timing and pattern-information analyses of fMRI data

Abstract Pattern-information approaches to fMRI data analysis are becoming increasingly popular but few studies to date have investigated experimental design optimization for these analyses. Here, we tested several designs that varied in the number of trials and trial timing within fixed duration scans while participants encoded images of animals and tools. Trial timing conditions with fixed onset-to-onset timing ranged from slow 12-s trials with two repetitions of each item to quick 6-s trials with four repetitions per item. We also tested a jittered version of the quick design with 4–8 s trials. We assessed the effect of trial timing on three dependent measures: category-level (animals vs. tools) decoding accuracy using a multivoxel pattern analysis, item-level (e.g., cat vs. dog vs. lion) information estimates using pattern similarity analysis, and memory effects comparing pattern similarity scores across repetitions of individual items subsequently remembered vs. forgotten. For single trial estimates, category decoding was equal across all trial timing conditions while item-level information and memory effects were better detected using slow trial timing. When modeling events on an item-by-item basis across all repetitions of a given item, a larger number of quick, regularly spaced trials provided an advantage over fewer slow trials for category decoding while item-level information was comparable across conditions. Jittered and non-jittered versions of the quick trial timing did not differ significantly in any analysis. These results will help inform experimental design choices in future studies planning to employ pattern-information analyses and demonstrate that design optimization guidelines developed for univariate analyses of a few conditions are not necessarily optimal for pattern-information analyses and condition-rich designs.