Only 5% of excitatory input to primary visual cortical (V1) neurons corresponds to feedforward input from the retina, and only 20% of responses by these neurons can be explained by retinal input (Carandini, 2005; Muckli and Petro, 2013). Neuronal responses are therefore highly influenced by non-feedforward interactions, allowing the brain to combine external input from the retina with context and knowledge. This is accomplished by integrating feedforward input with signals from neurons processing higher-level or associative information. The signals transmitted from higher cortical areas to V1 are known as cortical feedback. The neuroscientific community is in agreement that cortical feedback is an important aspect of brain processing. However, the information transmitted by feedback and what factors give rise to contextual feedback remain largely unknown. Feedback connections provide V1 neurons with information about their far surround receptive fields (Angelucci and Bressloff, 2006), and stimulation in the surround provides contextual information about the scene to non-stimulated portions of V1 (Smith and Muckli, 2010; Muckli et al., 2015). Occluded V1 activity patterns recorded using fMRI have been used to decode different scenes, but again, little is known regarding the nature and content of the contextual information that feedback transmits. This thesis aims to examine the information in contextual feedback to early visual cortex, with a particular focus on V1. To investigate this topic we used an occlusion paradigm derived from that of Smith and Muckli (2010) and Muckli et al. (2015). During normal vision, both feedforward and feedback signals are present. As such, a useful approach to study feedback is to isolate it from feedforward input. We occluded one quadrant of the visual field during stimulus presentation in order to remove meaningful feedforward input about scenes in a portion of retinotopic visual cortex. We used fMRI to assess brain activity in early visual cortex, allowing us to detect dendritic signaling associated with cortical feedback due to its sensitivity to cortical energy consumption (Logothetis, 2007, 2008; Petro et al., 2014). In Chapter 2, we investigated potential high-level information in cortical feedback to V1 and V2. We presented subjects with an expanded version of the occlusion paradigm from Smith and Muckli (2010) and Muckli et al. (2015). We included twenty-four partially occluded scenes from six categories and spatial depths. These two high-level scene characteristics were chosen because they have previously been shown to modulate early visual cortical responses (Walther et al., 2009; Kravitz et al., 2011). We were therefore interested in whether these characteristics also modulate feedback to V1 and V2. We found that response patterns in these subregions contain high-level category information, but we did not find that visual depth information generalized across exemplars. Additionally, we found that retinotopic responses in Occluded V1 and V2 differed from each other, suggesting that feedback to these two areas has different information content, and matching the known anatomical connections from mid- and higher-level visual areas in the ventral stream (Rockland et al., 1994; Rockland and Ojima, 2003). In Chapter 3, we probed the information content of Occluded V1 and V2 responses at multiple levels of complexity using Representational Similarity Analysis (RSA) and encoding models. By analyzing data from Chapter 2 in these frameworks, we were able to compare both local (voxelwise) and distributed (multi-voxel) Occluded responses to three biologically-inspired computational models (the contrast energy-based Weibull model, the orientation-based Gist model, and the mid-level vision H-Max model), and the high-level scene characteristics explored in Chapter 2. Using RSA, we also compared scene representations from Occluded and Non-Occluded areas. We found that in Non-Occluded areas, V1 and V2 represent scenes similarly, while Occluded V1 and V2 do not. We also found that scene representations in Occluded V1 and V2 were correlated with high-level Category and H-Max models. Individual voxel encoding models showed that Occluded V1 voxels within 5◦ visual angle of fixation encode low-level information about the occluded scene, while voxels outside of 5◦ encode higher-level information. These results highlight a potential visual field bias in the type of information transmitted to V1 through feedback, with foveal voxels receiving more precise, low-level scene information, and peripheral voxels receiving more invariant or global scene features. In Chapter 4, we examined the laminar profile of Occluded V1 using high-resolution (0.8mm3) 7T fMRI. We again expanded our stimulus set, now with 192 Occluded scenes and 192 Non-Occluded scenes. This large stimulus set allowed us to map scene information onto voxel responses in greater detail, and the use of both Occluded and Non-Occluded scenes allowed us to compare voxel responses when receiving only feedback with responses when receiving feedforward, lateral and feedback information. We found that V1 responses exhibit predictive and high-level response properties in addition to feedforward orientation and spatial frequency properties typically associated with V1 responses. These predictive and high-level responses were primarily associated with superficial layers of cortex. We also found that voxel tuning toward feedforward and feedback signals was different between cortical layers of V1. Our findings suggest that feedback connections terminating in superficial layers provide V1 neurons with contextual and associative information not available via localized feedforward input. The neuroscientific results presented in this thesis extend our knowledge about the information content of cortical feedback to early visual cortex. These results add support to the notion that V1 can be considered to speak two languages (Muckli and Petro, 2013). Not only does it play a role as an early stage of processing of sensory visual input, where it deals with processing low-level features, but it also receives messages from diverse areas of cortex and these messages supplement local processing by providing contextual information.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:754360 |
Date | January 2018 |
Creators | Morgan, Andrew T. |
Publisher | University of Glasgow |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://theses.gla.ac.uk/30713/ |
Page generated in 0.0018 seconds