A comparison of methods for characterizing the event-related BOLD timeseries in rapid fMRI

Abstract

Information about the shape and temporal duration of the blood oxygenation level dependent (BOLD) response can inform both functional neuroanatomy and psychological theory. However, the BOLD response evolves over 20 s or more, making it difficult to distinguish the unique characteristics of the response evoked by temporally adjacent stimuli. Fortunately, event-related BOLD signals can be extracted given that there is adequate variance in the distribution of inter-stimulus intervals (ISI). Unfortunately, the ISI distribution that yields the highest statistical efficiency is not always optimal from a psychological perspective; variability in the stimulus timing may complicate the interpretation of neuroimaging data in terms of underlying cognitive operations. In the present paper, Monte Carlo simulations are used to evaluate two techniques for estimating the event-related BOLD timeseries—event-related averaging and deconvolution using the Ordinary Least Squares estimate —with respect to maintaining acceptable levels of statistical power and experimental validity. While the unbiased deconvolution technique more robustly estimates the shape of the BOLD response functions, both methods succeed in accurately re-producing known differences between evoked BOLD responses when the stimulus ordering is randomized. However, the deconvolution method is more effective at preserving differences when there are sequential dependencies in the stimulus presentation order and restricted ISI distributions are used; particularly if the second of two sequentially dependent stimuli is omitted on some portion of the trials. Importantly, the successful re-production of the evoked BOLD response using restricted ISI distributions often maximizes the ability to make psychologically valid experimental conclusions.