Abstract
Block designs are often used with multivariate pattern analyses (MVPA) of fMRI studies to achieve high classification accuracy rather than fast event-related designs which suffer from low accuracy. This lower accuracy is typically attributed to the interaction between the rapid succession of stimulus events and the temporally sluggish BOLD response. We show that much of the higher classification accuracy derived from the use of block designs is a result of the averaging of fMRI data points during analysis rather than the different temporal characteristics of the two designs. Observers judged the orientation of centrally presented Gabor patches (10° diameter, 1.2 cpd) oriented at +/-45° relative to the vertical. Trial durations were 2 seconds with an initial 200 ms presentation of the Gabor stimulus. Experimental runs consisted of 25 presentations of each condition including 25 fixation trials of equal duration. Observers participated in three scanning sessions which differed only in the ordering of experimental trials: either blocked or fast event-related (m-sequence or genetic algorithm optimized designs). Each observer completed 10 runs per session. We used a linear SVM to assess the orientation discrimination accuracy of retinotopically defined visual cortex in two ways: training and predicting on single trials (fMRI data points shifted by 4 seconds) or trials averaged across blocks. Trials from event-related designs were grouped into three blocks matching the block design. Averaging trials produced a significantly higher classification accuracy than single trial analysis for all experimental designs. Further, the single trial analysis accuracy of the block design was close to chance across visual areas with no significant accuracy difference between designs. Our results suggest that much of the benefit that block designs provide in MVPA fMRI studies is due to the averaging of fMRI data during analysis and this technique can equally well be applied to fast event-related designs.
Army grant W911NF- 09-D-0001.