Estimating Neural Networks with State-of-the-Art Neuronal Computational Models

Structural, functional, and effective network models have been developed to estimate the organization in the brain from a wide range of spatial dimensions and neuroimaging methods. Novel neurophysiological recording technology that helps investigate brain function from new perspectives has recently become available. Micro-Electrocorticograph (Micro-ECoG) arrays offer the flexibility to record electrical activity from the surface of the cortex from high density electrodes that are sub-mm in diameter. It remains unclear what information these arrays can provide about the underlying signal generation and what neurophysiological dimensions are relevant. The columnar organization hypothesis is currently the most widely adopted to explain the cortical processing of information. Analyzing microECoG potentials with neuronal computational models could uncover relevant information regarding the underlying signal generation. Volume conduction model attempts to model the activity with electrostatic representation from a single cortical generator. Dynamic causal model (DCM) is a generative model comprised of cortical columns connected in a network fashion.;We showed that activity located < 1000 microns can be described by the single source volume conduction model with reasonable accuracy but distances any greater than 1500 microns require a network model, like the connected network DCM, to accurately predict the potentials observed. The estimated sensory network model for two cortical columns, confirmed by other neurophysiological models, illustrate the local organization of the sensory cortex and provide likely signal generators for the LFPs recorded from the cutting-edge microECoG array. Multiple new computational methods were developed as a result and are discussed throughout. Quantitative metrics were developed to assess model fidelity and make objective comparisons across different models. We are shedding light onto an important topic no one has explored before. We can estimate a network between two cortical columns from electrical activity recorded without damaging the brain. These network estimates serve as a model going forward that can be tested under different conditions.;Two key assumptions were made about the location of the stimulus input for the sensory evoked experiment in the Chapter 4. First, that the input was relayed into the spiny stellate layer of the modeled cortical column. We adapted DCM in order to model the sensory evoked potentials with the input located in three different layers: spiny stellate, superficial pyramidal, and the deep pyramidal. The results of this experiment confirm the naturalistic assumption the relayed input innervates the spiny stellate layer of the modeled cortical column. In addition, the log-evidence distribution generated under a bootstrap method, provides a metric to compare different DCM models and the differences that can be expected under the three different input scenarios. The second assumption was that the input was relayed into the cortical column located directly under the primary channel. Optogenetics technology can provide the means to resolve these assumptions and provides an alternative way to stimulate a given cortical network with greater specificity and precision. We employed DCM with the CMC variation with the three different input scenarios to accurately model microECoG potentials evoked by the optogenetic penetrating fiber stimulation. The deep pyramidal input DCM outperformed alternative input locations in predicting activity when the fiber was located at the surface of the cortex, disagreeing with our intuition. However, this result can be well explained by modeling efforts explaining the apical dendrites and the soma of the pyramidal cell are the two most light-sensitive components. As a result, we can successfully model sensory evoked and optogenetic evoked microECoG recorded potentials with the input in different layers in order to test the relevance of the input location under different scenarios. This provides a tool that can be used to shed light into the location of the input into a DCM estimated network from datasets using novel stimulation techniques.;The current standard of care for seizure localization incorporates numerous noninvasive and minimally invasive neuroimaging methods. Yet their ability to characterize the epileptogenic network is limited. We evaluate a novel data-driven technique that has been used to detect fMRI signal changes in epilepsy patients that may be congruent with interictal epileptiform discharges (IED) known as Temporal Clustering Analysis (TCA). The location of the nearest cluster from these activation maps varied in proximity to the area of seizure onset determined clinically. The approach adopted has the potential to be used clinically as it is non-invasive like EEG, computationally quick to implement, and can be used in patients with different levels of disease severity, and functional status.