Document Type

Article

Publication Date

5-23-2014

Publication Title

Frontiers in Neuroscience

Department

Department of Psychological and Brain Sciences

Abstract

Perception is an active process that interprets and structures the stimulus input based on assumptions about its possible causes. We use real-time functional magnetic resonance imaging (rtfMRI) to investigate a particularly powerful demonstration of dynamic object integration in which the same physical stimulus intermittently elicits categorically different conscious object percepts. In this study, we simulated an outline object that is moving behind a narrow slit. With such displays, the physically identical stimulus can elicit categorically different percepts that either correspond closely to the physical stimulus (vertically moving line segments) or represent a hypothesis about the underlying cause of the physical stimulus (a horizontally moving object that is partly occluded). In the latter case, the brain must construct an object from the input sequence. Combining rtfMRI with machine learning techniques we show that it is possible to determine online the momentary state of a subject’s conscious percept from time resolved BOLD-activity. In addition, we found that feedback about the currently decoded percept increased the decoding rates compared to prior fMRI recordings of the same stimulus without feedback presentation. The analysis of the trained classifier revealed a brain network that discriminates contents of conscious perception with antagonistic interactions between early sensory areas that represent physical stimulus properties and higher-tier brain areas. During integrated object percepts, brain activity decreases in early sensory areas and increases in higher-tier areas. We conclude that it is possible to use BOLD responses to reliably track the contents of conscious visual perception with a relatively high temporal resolution. We suggest that our approach can also be used to investigate the neural basis of auditory object formation and discuss the results in the context of predictive coding theory.

DOI

10.3389/fnins.2014.00116

Share

COinS