Design and statistical analysis of fMRI experiments to assess human brain hemodynamic responses

Abstract

The vast majority of previous functional magnetic resonance imaging (fMRI) studies have used simple 'block' experimental designs. 'Block' designs are those designs in which the same stimulus is presented to the subject over a relatively long period of time. These studies are limited in their ability to probe brain function in that they only explore steady-state differences, are confounded by cognitive babituation effects, and cannot be compared to traditional behavioral and electrophysiological experiments. Event­related (brief ,timulus presentation) techniques of electrophysiological experiments have recently been applied to fMRI, although most efforts have been far from optimal. The overall goal of this work was to develop efficient and robust techniques to estimate brain activity for event-related fMR.I experiments. We first performed two experiments to assess the steady-state linearity of the hemodynamic system in prinwy visual cortex (Vl) for an event-related visual stimuli. In agreement with previous studies, we found that the system was approximately linear. Given this result, we used linear estimation techniques to estimate the hemodynamic response during rapid, event-related experiments using two different design strategies. We found that designs using a geometric distribution of presentation intervals were insensitive to nonlinearities, and that these designs enabled very rapid presentation experiments. We then developed a general statistical hypothesis framework to test for activated brain regions during event-related experiments. In particular, we made spatially local and global estimates of the underlying physiological noise process. The sensitivity and specificity of three different hypothesis tests were validated with synthetic noise, actual fMRI noise, actual fMR.I noise plus synthetic activation, and actual fMR.I activation and noise data. The overall work allows for more efficient and appropriate fMR.I response detection and potentially new classes of fMRI experiments.