Analysis of Neuronal Source Dynamics During Seizure Using Vector Autoregressive Models, ICA, Sparse Bayesian Learning and ECoG

Accurate detection of seizure onset as well as identification of neuronal regions critically involved in initiating and propagating a seizure remains an important area of research. Understanding the dynamics of neural processes underlying different stages of a seizure can help in devising novel methods of seizure detection, intervention and treatment. In this paper we analyze linear neuronal dynamics during epileptic seizures using adaptive multivariate autoregressive (VAR) models applied to maximally-independent (ICA) sources of intracranial EEG (iEEG, ECoG) data recorded from subdural electrodes implanted in a human patient for presurgery monitoring. We analyze the time-frequency dynamics of directed information flow between sources using a multivariate granger-causal method (dDTF), identifying distinct patterns of information flow in different stages of the seizure. We then further examine the spatial distribution in the cortical source domain of causal sources and sinks of ictal activity using a novel combination of causal flow metrics and Sparse Bayesian Learning-based source localization. Finally, we apply an eigendecomposition method to decompose the VAR model into a system of decoupled oscillators and relaxators (eigenmodes) with characteristic damping times and frequencies. We demonstrate that analysis of a small subset of the most dynamically important eigenmodes may allow effective detection of ictal onset and offset, while also yeilding insight into the dynamical structure of the neuronal system.