The role of long-range connections in contextual processing and spontaneous activity of primary visual cortex

The aim of this work is to set the basis for the development of a theoretical framework to investigate how artificial signals can be successfully introduced into primary visual cortex through electrical stimulation. This goal is approached by focusing on two different aspects of visual information processing: the contextual modulations that occur when localized visual stimuli are placed in conjunction with surround stimuli and the spontaneous activity that emerges in the absence of sensory stimulation.
Generalizing the well known standard sparse coding framework, we propose a generative model to encode spatially extended visual scenes. We show that pairing an anatomically inspired constraint (which imposes that neurons have direct access only to small portions of the visual field) to a computational coding principle (whose goal is to maximize accuracy and sparseness of stimuli-representation) is sufficient to account for a number of heterogeneous features. In particular, when trained with natural images, the model predicts a connectivity structure linking neurons with similar orientation preferences matching the typical patterns found for long-ranging horizontal axons and feedback projections in visual cortex. When subjected to contextual stimuli typically used in empirical studies, it replicates several hallmark effects of surround modulation, some of which previously unexplained, and provides a uniform explanation to contextual processing.
The dynamics of ongoing activity in primary visual cortex was investigated in a structurally simple model, where the network connectivity was chosen to mimic what we obtained from the optimization process in the sparse coding model. We used both analytical and numerical methods to study the patterns of activity that the model exhibited, identifying conditions under which biophysically realistic orientation-tuned states emerged. We quantified several properties important for comparing the model to experimental data, such as the emergence and decay probability, average persistence, localization and coexistence of different states.
In both studies, we show to what extent the properties of long-range connections between visual cortical neurons are responsible for the observed empirical facts, proposing a well- defined functional role for horizontal axons and feedback projections for contextual processing phenomena and for the generation of spontaneous tuned states.
In the last part of this thesis, we tackle more concretely the problem of inducing artificial perceptions via electrical stimulation of primary visual cortex. We present a new stimulation- paradigm which consists in monitoring the spontaneous orientation-tuned states and delivering a weak modulatory current when the cortex is in a desired state, to induce spikes in neurons that are currently close to their firing threshold. The proposed framework is tested in a structurally simple spiking neural network whose activity resembles spontaneous activity in V1. After calibrating the model to a physiologically realistic operating point, we conduct a feasibility study, investigating in particular the relations between stimulation amplitude, temporal resolution and specificity of the percept. We then show how this strategy has the potential to result in the artificial perception of an image composed by a combination of oriented features, an improvement with respect to the round phosphenes typically observed in experiments.