Abstract
It is commonly assumed that event-related functional MRI studies of the visual system require slow stimulus presentation rates, with stimuli typically presented every 3-5s. At this rate, the BOLD signal is expected to be linear and can therefore be analyzed in a linear modeling framework. However, this assumption conflicts with recent findings that have successfully mapped the content of video stimuli to brain responses and with other work using more rapidly changing stimuli, suggesting that higher presentation rates may be possible. To address these seemingly conflicting views, we used simulations and measured brain responses with 7 Tesla fMRI (TR = 500ms) to determine the extent to which rapid stimulus presentation is achievable. For fMRI, we varied presentation rates between 0.5s and 4s and presented observers with images of faces, places, objects, and scrambled objects. Our simulations showed that design efficiency increases with higher stimulus presentation rates under two conditions: (1) when a large number of stimulus conditions is used (also known as condition-rich designs), or (2) when the contrast between stimuli - the critical measure for most fMRI studies - is the focus, rather than estimating conditions independently. For fMRI, we were able to identify the category-selective brain regions FFA, PPA, and LOC, and we observed a graded response, with increased effects for faster presentation rates. Using multivariate pattern analysis, our results also showed that the larger number of stimulus repetitions leads to better alignment of object representations derived from BOLD signals with those from human behavioral data and artificial neural networks. Together, our results demonstrate the usefulness of higher stimulus presentation rates, specifically for condition-rich, contrastive fMRI study designs. This allows either reducing scanning time or increasing the discriminability of stimulus effects, depending on the study goals.