Abstract
The information received from our senses is typically consistent with a range of possible stimulus values, making any of our perceptual decisions uncertain. It is well known that this perceptual uncertainty affects behavior, but how does the brain represent knowledge of uncertainty? We used functional MRI in combination with pattern-based analyses to address this question. Participants viewed annular gratings of random orientation. Shortly after the presentation of the grating, observers reported its orientation. We used a novel pattern-based decoding approach to analyzing fMRI data, computing for each trial of BOLD activity the likelihood function of stimulus orientation. This approach differs from conventional decoding approaches in that the latter typically generates a single orientation estimate, whereas our method computes a full probability distribution over all possible orientations. We hypothesized that the width of the decoded probability distribution might reflect the degree of perceptual uncertainty. Accordingly, we compared the width of this distribution with behavioral variability, reasoning that more precise stimulus representations in visual cortex should be linked to less variable (more accurate) behavior. We found that perceptual uncertainty could reliably be decoded from fMRI activity patterns in the visual cortex. Specifically, the trial-to-trial fluctuations in the width of the likelihood function reliably predicted the variability in the observer's response. In contrast, we observed no link between the overall amplitude of the BOLD response and behavioral performance. These findings provide some of the first evidence that the trial-by-trial precision of perception can reliably be extracted from the human visual cortex.
Meeting abstract presented at VSS 2014