Nonlinear Kalman-filtering estimation of unobserved cortical layer population activity from EEG recordings in a realistic cortical micro-architecture model

Abstract

The Electroencephalogram (EEG) measures global brain electrical activity using electrodes placed on the scalp. EEG measurement continues to be an important tool for understanding brain dynamics since it is closely related to the underlying neuronal activity in real time. A main challenge in relating EEG to its neural origin is the non-uniqueness of possible sources (or so called inverse solution). In this regard, constraining the estimation problem using realistic models of EEG generation can produce a small set of potential source configurations. In this thesis we aim to, first, introduce a set of neuro-physiologically plausible models that simulate EEG generation and, second, to employ novel nonlinear estimation tools based on Kalman filtering to identify, within the developed models, a set of key parameters that affect EEG generation under specific cognitive states. Cubature kalman filter (CKF), a state of art nonlinear estimator developed in 2009 and recently applied in neuroscience on understanding fMRI signals, is used for the first time for estimation based on EEG signals in order to reveal contributing neural brain dynamics and localize current sources. The thesis presents simulation and estimation results based on a case scenario of idling and attentive processing in the early and intermediate visual cortical areas (V1-V4, IT) relying on recent literature describing intra-cortical origins of alpha oscillations collected in awake behaving monkeys. Overall, CKF estimation shows accurate prediction of hidden neuronal firing in the simulated model despite the highly nonlinear interaction between cortical populations within lamina and across hierarchies in the visual cortex.