Analysis and inference of brain links using information theory and non-linear time-series analysis

Abstract

The selection of a model of the brain is of practical interest for the understanding of processes such as the sensorial perception, diseases such as Epilepsy, and the reaction of the brain to rapid loss of neurons, such as due to Alzheimer's or following a brain stroke. However, standard methods of analysis are not adequate to the problem due to the high complexity of the brain connections and the neuronal dynamics. Hence, the understanding of the brain system requires the analysis of the features measured by EEG/MEG/fMRI/SPECT, as well as the adoption of tools of inference of direct or functional links able to deal with the high complexity of the links of the components. Duarte et al. (2017) considered the problem of inference of the interaction graph of neurons in a stochastic model. Such option for a neuron network model, for which NeuroMat is the international reference centre, proven successful for analysis of brain reaction to external stimuli (Duarte et al., 2018). This research aims to contribute to the further development of the inferential methods for the stochastic brain model and associate with tools from information theory to the analysis of EEG signals, with particular emphasis to cases with applicability to support the diagnosis of diseases. In case of possible extension, we aim to develop a tool for use in diagnostic centres of the Brazilian National Health Service. This research contributes to the NeuroMat project by assisting in the mathematical development of the stochastic brain model, however focusing more on the aspects of data-based inference. This research inserts in the NeuroMat project research line 'inference of the function interaction between neuron structures'. A. Duarte, A. Galves, E. Locherbach and G. Ost. Estimating the interaction graph of stochastic neural dynamics. Preprint, 2017. https://arxiv.org/pdf/1604.00419.pdfA. Duarte, R. Fraiman, A. Galves, G. Ost and C. D. Vargas. Context tree selection for functional data. Preprint, 2018. https://arxiv.org/pdf/1602.00579.pdf (AU)