Transfer entropy—a model-free measure of effective connectivity for the neurosciences

Max Planck Institute for Brain Research, Frankfurt, Germany.
Journal of Computational Neuroscience (Impact Factor: 1.74). 02/2011; 30(1):45-67. DOI: 10.1007/s10827-010-0262-3
Source: PubMed

ABSTRACT Understanding causal relationships, or effective connectivity, between parts of the brain is of utmost importance because a large part of the brain's activity is thought to be internally generated and, hence, quantifying stimulus response relationships alone does not fully describe brain dynamics. Past efforts to determine effective connectivity mostly relied on model based approaches such as Granger causality or dynamic causal modeling. Transfer entropy (TE) is an alternative measure of effective connectivity based on information theory. TE does not require a model of the interaction and is inherently non-linear. We investigated the applicability of TE as a metric in a test for effective connectivity to electrophysiological data based on simulations and magnetoencephalography (MEG) recordings in a simple motor task. In particular, we demonstrate that TE improved the detectability of effective connectivity for non-linear interactions, and for sensor level MEG signals where linear methods are hampered by signal-cross-talk due to volume conduction.

Download full-text


Available from: Gordon Pipa, Sep 29, 2015
1 Follower
39 Reads
  • Source
    • "Moreover, using this method, we found the transfer of information between two variables in both directions, i.e. from maternal to fetal heart rate and vice versa. TE has been used for investigating the coupling of physiological variables in various applications [10]–[12]. Improved methods and toolboxes for TE estimation have been recently proposed [13], [14]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Although evidence of the short term relationship between maternal and fetal heart rates has been found in previous model-based studies, knowledge about the mechanism and patterns of the coupling during gestation is still limited. In this study, a model-free method based on Transfer Entropy (TE) was applied to quantify the maternal-fetal heart rate couplings in both directions. Furthermore, analysis of the lag at which TE was maximum and its changes throughout gestation, provided more information about the mechanism of coupling and its latency. Experimental results based on fetal electrocardiograms (fECGs) and maternal ECG showed the evidence of coupling for 62 out of 65 healthy mothers and fetuses in each direction, by statistically validating against the surrogate pairs. The fetuses were divided into three gestational age groups: early (16-25 weeks), mid (26-31 weeks) and late (32-41 weeks) gestation. The maximum TE from maternal to fetal heart rate significantly increased from early to mid gestation, while the coupling delay on both directions decreased significantly from mid to late gestation. These changes occur concomitant with the maturation of the fetal sensory and autonomic nervous systems with advancing gestational age. In conclusion, the application of TE with delays revealed detailed information about the changes in fetal-maternal heart rate coupling strength and latency throughout gestation, which could provide novel clinical markers of fetal development and well-being.
    IEEE Engineering in Medicine and Biology Conference (EMBC); 08/2015
  • Source
    • "Again, data cleaning data from bad components using methods such as ICA can be beneficial to obtain cleaner data; and (3) low cross-talk between the measurements of network entities. To tackle this issue, which arises mostly because of the scalp volume conductance , GC analyses can be performed by applying GC on ICA brain sources or current source density (also known as Spatial Laplacians) signals instead of original brain voltages from scalp channels (Vicente et al., 2011; Coben et al., 2014). "
    [Show abstract] [Hide abstract]
    ABSTRACT: Epilepsy is a chronic neurological disorder characterized by repeated seizures or excessive electrical discharges in a group of brain cells. Prevalence rates include about 50 million people worldwide and 10% of all people have at least one seizure at one time in their lives. Connectivity models of epilepsy serve to provide a deeper understanding of the processes that control and regulate seizure activity. These models have received initial support and have included measures of EEG, MEG, and MRI connectivity. Preliminary findings have shown regions of increased connectivity in the immediate regions surrounding the seizure foci and associated low connectivity in nearby regions and pathways. There is also early evidence to suggest that these patterns change during ictal events and that these changes may even by related to the occurrence or triggering of seizure events. We present data showing how Granger causality can be used with EEG data to measure connectivity across brain regions involved in ictal events and their resolution. We have provided two case examples as a demonstration of how to obtain and interpret such data. EEG data of ictal events are processed, converted to independent components and their dipole localizations, and these are used to measure causality and connectivity between these locations. Both examples have shown hypercoupling near the seizure foci and low causality across nearby and associated neuronal pathways. This technique also allows us to track how these measures change over time and during the ictal and post-ictal periods. Areas for further research into this technique, its application to epilepsy, and the formation of more effective therapeutic interventions are recommended.
    Frontiers in Human Neuroscience 07/2015; 9:194. DOI:10.3389/fnhum.2015.00194 · 2.99 Impact Factor
    • "However, this procedure may complicate the interpretation, as what is being assessed is the causal connectivity among changes in each time series. New approaches such as adaptive MVAR models (Hesse et al., 2003; Astolfi et al., 2008) that make no assumptions about the stationarity of the signals or information theoretic tools, such as transfer entropy (Schreiber, 2000), a model-free method that measures directed non-linear and linear information flow have been proposed and applied successfully to simulated and real electrophysiological data (Vicente et al., 2011; Plomp et al., 2014). Future studies with new statistical approaches should attempt to determine the effective functional mechanisms that underlie the observed data. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Magnetoencephalography was recorded during a matching-to-sample plus cueing paradigm, in which participants judged the occurrence of changes in either categorical (CAT) or coordinate (COO) spatial relations. Previously, parietal and frontal lobes were identified as key areas in processing spatial relations and it was shown that each hemisphere was differently involved and modulated by the scope of the attention window (e.g. a large and small cue). In this study, Granger analysis highlighted the patterns of causality among involved brain areas - the direction of information transfer ran from the frontal to the visual cortex in the right hemisphere, whereas it ran in the opposite direction in the left side. Thus, the right frontal area seems to exert top-down influence, supporting the idea that, in this task, top-down signals are selectively related to the right side. Additionally, for CAT change preceded by a small cue, the right frontal gyrus was not involved in the information transfer, indicating a selective specialization of the left hemisphere for this condition. The present findings strengthen the conclusion of the presence of a remarkable hemispheric specialization for spatial relation processing and illustrate the complex interactions between the lateralized parts of the neural network. Moreover, they illustrate how focusing attention over large or small regions of the visual field engages these lateralized networks differently, particularly in the frontal regions of each hemisphere, consistent with the theory that spatial relation judgements require a fronto-parietal network in the left hemisphere for categorical relations and on the right hemisphere for coordinate spatial processing. © 2015 Federation of European Neuroscience Societies and John Wiley & Sons Ltd.
    European Journal of Neuroscience 02/2015; DOI:10.1111/ejn.12846 · 3.18 Impact Factor
Show more