Natural image coding in early visual areas: Functional magnetic resonance imaging and psychophysical studies of the human visual system.

摘要

This work examines visual responses to broadband images, at the same time exploring the utility of high-field functional magnetic resonance imaging (fMRI) for studying neural population responses in early visual areas. The richness of natural images hinders quantitative studies of how local features combine to determine perception of the image as a whole, yet these interactions are crucial for understanding visual processing of natural images. Rather than studying responses to individual images, the statistical properties of sets of images are manipulated to study how contrast, spatial frequency, and spatial phase coherence interact to determine detection and discrimination thresholds for broadband images. As a non-invasive neuroimaging technique with an unlimited field of view and high spatial resolution, functional MRI provides a promising tool for imaging the response of populations of neurons to natural stimuli. However, the majority of fMRI is based on the blood oxygenation level-dependent (BOLD) response, a blood flow response only indirectly linked to neural activity. Whether or not BOLD fMRI is a useful tool for studying these neural response properties depends on our ability to quantify the relationship between neural activity and the BOLD response.; This work represents a first step toward a quantitative model of the spatial distribution of neural activity in V1 in response to natural images, as well as the interaction of neural activity and BOLD fMRI. The data presented here reveal the separate roles of spatial frequency, spatial phase coherence, and feature density in determining the visibility of grayscale images. The resulting conclusion is that, with a careful understanding of the BOLD fMRI signal source and the link to underlying neural activity, the BOLD signal contains information about neural population activity that would otherwise be inaccessible. Extension of this working model to include the temporal dynamics of the BOLD responses, as well as more detailed information about the spatial distribution of activity in V1, will form the basis for continued studies of visual response to specific image features and the interactions of spatially distributed neural responses within and between visual areas.