The predicting brain: anticipation of moving objects in human visual cortex

Abstract

The human brain is nearly constantly subjected to visual motion signals originating from a large variety of external sources. It is the job of the central nervous system to determine correspondence among visual motion input across spatially distant locations within certain time frames. In order to create stable and coherent percepts of the visual scenery, the brain is required to integrate visual motion signals over space and time, which may be accomplished by creating predictions of upcoming visual input. A recent fMRI BOLD study has shown that integration of visual motion input coincides with biased BOLD activity for radial motion directions at both peripheral and foveal cortical visual field representations. The current thesis, investigates mechanisms of visual motion integration through the presence or absence of motion biases using high-field 7 Tesla BOLD fMRI. We found that motion biases only emerged for motion stimuli that exhibited an uninterrupted motion trajectory during the perception of random dot kinematograms. More specifically, motion biases emerged at the onset of motion trajectories, which reflects different responses to novel appearing moving dots versus dots that were detected previously by other neurons. Thus, the visual system plausibly uses predictions to encode visual motion input. Accordingly, we show in several following experiments that the amplitude of the BOLD response depends on the novelty of the visual motion input regardless of location in the visual field and motion direction. BOLD signals to various motion stimuli were always enhanced at the onset of motion stimuli, where appearance of motion signals could not be predicted, while gradually decreasing for motion signals towards the end of motion trajectories. These effects were found for both random dot kinematograms and single moving objects, and appear to originate at the lowest levels of visual motion processing. We also managed to exclude several possible alternative explanations for present findings, such as classical receptive field effects, shifts in spatial attention towards novel motion, flexible retinotopy, and blurring of motion signals due to low temporal integration. In the current thesis, we show that the BOLD response to visual motion plausibly reflects prediction mechanisms used by the visual system during visual motion processing in human early visual cortex. However, the mechanisms at work might not literally predict each possible change within the visual field. Predictions of visual motion input may instead be heuristically determined by means of an automatic suppression which is conveyed through horizontal connections in the direction of visual motion. A simple suppression mechanism that allows for motion anticipation could be sufficiently effective to determine correspondence across motion signals, while being most energy efficient.