Summary: | Perceptual decision making refers to the process of making a choice among a series of options based on sensory information. Several studies have used visual stimuli to gain an understanding of the processes involved in encoding sensory information and its decoding, leading to a perceptual decision. One popular visual modality for studying these questions is motion and the ability to discriminate between axes of motion. Several mathematical models describing the processes of perceptual decision making have been proposed – many of them are based on data from electrophysiological experiments on macaque monkeys. By directly recording neuronal activity while monkeys were presented with different visual stimuli and making categorical choices about the perceived direction of motion, scientists have been able to study how decisions are made when enough perceptual evidence is accumulated to reach a threshold. A particularly interesting aspect of perceptual decision making is that it allows the study of situations in which the choice deviates from the physical features characterising the stimulus (e.g., a leftward motion is presented but the subject reports perceiving a rightward motion). A type of such perceptual bias is called reference repulsion: a systematic bias away from a reference when estimating the direction of motion of a stimulus. Several possible explanations of this phenomenon have been proposed: incorrect encoding of sensory information, influence of prior knowledge about the world, response-related factors such as expectations, rewards, and response history. The aim of this thesis was to shed light on when in the sequence of decision making such perceptual biases arise, as well as further address both sensory and higher-order factors that influence perceptual decisions of visual stimuli. We combined a series of psychophysical, eye-tracking, and neuroimaging studies, together with computational modelling approaches, to selectively look at the effect of: sensory information available during decision making, task-related sensory information processing, response modality, and also look for specific mechanisms involved in processing highly similar/dissimilar stimuli. The findings presented in this thesis show that perceptual biases in estimates of motion direction arise at a later stage than at the encoding of sensory representation, as previously thought. In particular, we show that information present at the time of the response is fundamental for the bias to emerge: the presence of a reference while estimating direction of motion results in reference repulsion, but this effect is not there when the same estimate is given in the absence of a reference. Moreover, the information given by the reference at the time of response – when subjects report the perceived motion, rather than at the time of stimulus presentation – plays a crucial role in the observed perceptual bias. These findings were used to develop a mathematical model able to describe the phenomena observed, as well as making a series of testable predictions. For example, the model could be used in future work to predict responses when more than one reference is present, when order of presentation of target and reference is inverted, and so on. By manipulating the modality with which subjects estimated the direction of motion of the stimuli they were presented with, it was also possible to show that a perceptual bias is observed for manual reproduction of the perceived direction, but not when the response is given by a saccadic eye movement. Finally, by looking at the brain activity recorded when performing a coarse / fine discrimination task in a functional magnetic resonance imaging (fMRI) study, we aimed at distinguishing between activity patterns encoding highly dissimilar / similar stimuli. For these analyses, we used both conventional, univariate analysis techniques, as well as a more advanced and relatively more recent multivariate approaches to the data. First, the retinotopic mapping of areas in early visual cortex and area MT was obtained through phase-encoded methods. Second, a version of the Generalised Linear Model was applied to the data measured while subjects were performing a fine / coarse discrimination task. This allowed to ensure the adequacy of tasks and stimuli used in the imaging study. I also applied the population Receptive Field methodology to fit a more explicit, physiologically relevant model of visual responses to the voxel-wise fMRI time series. Third, given that the spatial scale of the question we addressed in this study required aggregating sub-voxel differences in the fMRI responses during a fine versus coarse visual motion discrimination task, we employed a multivariate approach. This consisted in implementing a forward encoding model aimed at reducing the number of dimensions from several hundreds (given by the number of voxels) to a much smaller set of hypothetical channels. By considering the responses in these channels as a weighted combination from many hundred voxels we re-cast the activity patterns in a physiologically relevant space to predict responses to arbitrary visual motion directions. While there were very interesting aspects to the results from these imaging experiments, the analysis was inconclusive on any task-related shifts in stimulus encoding. Possible explanations, together with alternative paradigms that can be used in future to further address this question are discussed.
|