| Project Detail |
Towards a better model of vision
Drawing on expectations and context to make sense of the patterns of light entering the eyes is fundamental to our visual experience. Yet vision is usually modelled as a feedforward hierarchy. The EU-funded FFvsFB-UHF-fMRI project aims to create a better model by resolving the roles and mechanisms of feedforward and feedback processing. Using ultra-high field functional magnetic resonance imaging to monitor the distinct cortical layers receiving feedforward or feedback input, researchers will investigate how the processing streams interact in humans. They will use population receptive field mapping to characterise layer-specific receptive fields in visual crowding and Mooney image disambiguation – psychophysical paradigms in which spatial context and prior object knowledge determine perception. The goal is to understand how higher-level cortical areas shape low-level visual processing.
Visual perception has long been cast as an inference process, in which feedforward sensory signals are integrated with expectation-related feedback. For every feedforward sensory pathway, there is a reciprocal feedback projection, yet standard models continue to represent vision as a feedforward hierarchical network. Such models fail, however, when confronted with cases in which global spatial context, expectations, or other higher-order cognitive functions affect local processing. The characterization of the distinct roles and mechanisms of feedforward and feedback processing is thus crucial for better models of vision. Based on the theory of predictive coding, a proposed implementation of visual inference, we propose the receptive field (RF) as the mechanism by which feedforward and feedback processes interact. We present three projects that harness recent advances in neuroimaging techniques, allowing in vivo imaging of recurrent processing in humans. Ultra-high field (UHF) functional magnetic resonance imaging (fMRI) enables a hitherto impossible investigation of layer-specific cortical processing, with different cortical layers receiving input from either feedforward or feedback channels. We will leverage the sub-millimeter spatial resolution of UHF fMRI, together with population receptive field (pRF) mapping, to target feedforward and feedback RF properties in two behavioral paradigms - visual crowding and Mooney image disambiguation. Crowding is dependent on global spatial context, recruiting recurrent processing between early and mid-level visual areas, whereas the prior object knowledge manipulation involved in Mooney image disambiguation targets early to high visual areas. The investigation of the contextual modulation of layer-specific pRF properties will help elucidate the mechanisms by which higher cortical areas affect early sensory processing - a long-standing question in neuroscience, whose resolution is essential to the progress of vision research. |
