Analysis of causal relationship between different cortical regions measured by Transfer Entropy
Presenting author:
Despite the many advances in computational Neuroscience, to this day many aspects remain an enigma for scientists who have set out to unravel its function and structure. A general point of view is that of the information theory, which allows investigating the conditions and parameters involved in the information flow and processing for any type of system. We acknowledge the combination of this type of theory with neural dynamical models, driven by stimulus may provide a deeper understanding of transitions between brain states and inter-areal information flow in presence of complex cognitive paradigms.
In this work we make an analysis of the causality between brain regions in terms of Information Theory, we will use the information measure transfer entropy; this is the amount of directed transfer of information between two random processes X and Y. We propose make use of TIGRAMITE (Time Series Graph-Based Measures of Transfer entropy), this is a program that analyzes the transfer entropy of Time series and constructs graphs of causality in terms of information for the discovery of causal associations for the activity of different brain regions and at multiple time-lags, in this case from the time-series of simulated data of the behavior of the brain using the Neural Mass Model of Jansen and Rit, and real data of fMRI. We can also expand it to MEG and EEG.
The TIGRAMITE makes use of Conditional Mutual Information (CMI) to perform causal analysis of the discrete-time multivariate stochastic process. It adapts the PC algorithm for time series and makes some modifications to speed up the performance, using from the PC algorithm only the first step, the identification of the skeleton, and an independent test that the user selects, it can be a linear or nonlinear test, then, gives direction to the connections of the nodes using the Transfer Entropy. The program employs some interesting mechanisms to avoid infinite-dimensionality.
We proceed first with synthetic data of large neural mass models representing a spatially distributed network with several populations with oscillatory tuned to the different frequency bands. The data emulated the specific behavior of neural activity in three cortical layers as non-linear time series using Jansen and Rit Neural Mass Model. It represents the space-time organization of the brain and generates more complex dynamics between three structural levels (cortical unit, population, and system).
We also study fMRI data of the resting state of the Human Connectome Project (HCP) and focus on the visual regions to analyze the causal relationship between different structures of visual zones.
In this work we make an analysis of the causality between brain regions in terms of Information Theory, we will use the information measure transfer entropy; this is the amount of directed transfer of information between two random processes X and Y. We propose make use of TIGRAMITE (Time Series Graph-Based Measures of Transfer entropy), this is a program that analyzes the transfer entropy of Time series and constructs graphs of causality in terms of information for the discovery of causal associations for the activity of different brain regions and at multiple time-lags, in this case from the time-series of simulated data of the behavior of the brain using the Neural Mass Model of Jansen and Rit, and real data of fMRI. We can also expand it to MEG and EEG.
The TIGRAMITE makes use of Conditional Mutual Information (CMI) to perform causal analysis of the discrete-time multivariate stochastic process. It adapts the PC algorithm for time series and makes some modifications to speed up the performance, using from the PC algorithm only the first step, the identification of the skeleton, and an independent test that the user selects, it can be a linear or nonlinear test, then, gives direction to the connections of the nodes using the Transfer Entropy. The program employs some interesting mechanisms to avoid infinite-dimensionality.
We proceed first with synthetic data of large neural mass models representing a spatially distributed network with several populations with oscillatory tuned to the different frequency bands. The data emulated the specific behavior of neural activity in three cortical layers as non-linear time series using Jansen and Rit Neural Mass Model. It represents the space-time organization of the brain and generates more complex dynamics between three structural levels (cortical unit, population, and system).
We also study fMRI data of the resting state of the Human Connectome Project (HCP) and focus on the visual regions to analyze the causal relationship between different structures of visual zones.