Abstract. An important goal of cognitive neuroscience is understanding the neural underpinnings of conscious awareness. Although the low-level processing of sensory input is well understood in most modalities, it remains a challenge to understand how the brain translates such input into conscious awareness. Here, I argue that the application of multivariate pattern classification techniques to neuroimaging data acquired while observers experience perceptual illusions provides a unique way to dissociate sensory mechanisms from mechanisms underlying conscious awareness. Using this approach, it is possible to directly compare patterns of neural activity that correspond to the contents of awareness, independent from changes in sensory input, and to track these neural representations over time at high temporal resolution. I highlight five recent studies using this approach, and provide practical considerations and limitations for future implementations.
Keywords: perception, awareness, classification, decoding, illusionsOne of the important ambitions of research into perception is to identify the mechanisms involved in generating conscious awareness of stimulus properties. For most modalities, the physiological mechanisms underlying detection of a stimulus itself (in terms of converting stimulus energy into neural signals) are relatively well understood. What is much less well understood is how those sensory signals produce a sensory experience in our conscious awareness. This is currently the topic of considerable scientific interest.The advent of noninvasive neuroimaging techniques such as electro-and magnetoencephalography (EEG and MEG) and functional magnetic resonance imaging (f MRI) has given us powerful tools to investigate perceptual mechanisms in the human brain (for a review of multivariate neuroimaging approaches to the study of consciousness, see Haynes, 2009). However, isolating mechanisms related to the processing of low-level stimulus information from the mechanisms involved in realizing conscious awareness of that information is made difficult by the fact that the two are naturally highly correlated. Because it is difficult to influence what an observer perceives without also changing the actual stimulus presented to the observer, it is difficult to compare the neural signatures of two or more percepts without introducing differences in the processing of low-level information.Nevertheless, a number of paradigms have been devised to study changes in an observer's conscious percept without concurrent changes in the stimulus. One such approach is to use bistable stimuli, in which two (or more) stable perceptual interpretations are possible for a given stimulus. The Necker cube (Necker, 1832) and binocular rivalry are two classic examples. In this paradigm an observer views a static image, and the observer's percept switches between possible interpretations. This makes it possible to look for patterns of neural activity that correlate to each of the different interpretations. For example, Tong, Nakayama, ...
