Model-based exploration of interactions in the visual cortex

Abstract

The mammalian visual system is a dynamic and efficient data processing framework. Specifically, the cerebral cortex which is a structured network of cells benefits from an efficient information transmission coding. This thesis presents a model-based exploration of visual cortex in order to expand our current knowledge about brain mechanisms; specifically, the mechanism of information coding behind visual interactions. In the first study, we designed a functional magnetic resonance imaging (fMRI) experiment to explore a possible link between contextual modulation and efficient macroscopic spatial response coding in the visual cortex. The results imply that visual interactions were best explained with a decorrelation model which predicts average modulation strength by fully decorating the spatial fMRI signals. In the second study, we reviewed a potential approach to relate fMRI activation patterns to neural population activity. We went over existing knowledge about neurovascular coupling as a key point in predicting fMRI signal based on a neural network simulation and provided a sketch which covered practical steps to bridge the gap between mathematical modeling of single neuron responses to neuroimaging data with a mesoscopic biomimetic neural network. The proposed biomimetic neural network provides insight into data processing in cortical neural networks. In the third study, we designed an fMRI experiment and based on the blueprint of the second study simulated a simplified neural network representing the visual cortex. Then, we tried to replicate the experimental fMRI signal by means of this biophysically plausible neural network simulator. Our results highlight the role of dendritic structure of neurons to be able to repeat the experimental fMRI signals with high fidelity. In the fourth study, we used similar simulator as in the third study and tried to replicate expected neural activation pattern based on a well know contextual modulation (area summation function) in primary visual cortex. We anticipated that by getting closer to an activation pattern driven by area summation function, the efficiency of the neural network would be increased. Our results show that spiking frequency, entropy per spike and sparseness (as measures of network efficiency) are all associated with the natural area summation function. In summary, results of this thesis suggest that contextual modulation is related to efficiency of the visual system. In addition, it is possible to predict fMRI and expected area summation activation pattern by a mesoscopic neural network, however compartmental neurons have a key role to achieve this prediction.