Author: Hanieh Bakhshayesh
Bakhshayesh, Hanieh, 2019 Towards detecting connectivity in EEG: A comparative study of effective and functional connectivity measures on simulated data, Flinders University, College of Science and Engineering
Terms of Use: This electronic version is (or will be) made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. You may use this material for uses permitted under the Copyright Act 1968. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact copyright@flinders.edu.au with the details.
In neuroscience, there is considerable current interest in investigating the connections between
different parts of the brain. EEG is one modality for examining brain function, with advantages
such as high temporal resolution and low cost. Many measures of connectivity have been
proposed, but which is the best measure to use? In this thesis, we address the following
question: which measure is best able to detect connections between signals, in the challenging
situation of non-stationary and noisy data from nonlinear systems, like EEG?
The problem with using EEG is that we don’t known when a measure is giving the “right”
answer. Hence we choose to apply connectivity measures to simulated data that is similar to
EEG rather than EEG itself, so we always know the “right” answer.
We compare almost all of the most widely used or most promising measures, in total 26
functional connectivity measures and 20 effective connectivity measures. The performance of
functional connectivity measures is tested on simulated data from two systems: two coupled
Hénon maps; and two channels of simulated EEG. The performance of effective connectivity
measures is tested on simulated data from three systems: three coupled Hénon maps; a
multivariate autoregressive (MVAR) model with and without EEG as an exogenous input; and
simulated EEG. To determine whether connectivity is detected, surrogate data were generated
and analysed, and a threshold determined from the surrogate ensemble.
No measure performed best in all tested situations. In the comparison of the functional
connectivity measures, correlation and coherence performed best on stationary data with many
samples, in both high and low noise. S-estimator, correntropy coefficient, mean-phase
coherence (Hilbert), mutual information (kernel), nonlinear interdependence (S) and nonlinear
interdependence (N) performed most reliably on non-stationary data with small to medium
window sizes, in both high and low noise. Of these, correlation and S-estimator have execution
times that scale slower with the number of channels and the number of samples. In the
comparison of effective connectivity measures, the measures that model the data as MVAR
perform well when the data are drawn from that model. Frequency domain measures perform
well when the data have a clearly defined band of interest. When neither of these are true,
information theoretic measures perform well, as does Copula Granger causality.
Keywords: Biomedical signal processing,Connectivity, EEG,Nonstationarity
Subject: Science, Technology and Enterprise thesis
Thesis type: Doctor of Philosophy
Completed: 2019
School: College of Science and Engineering
Supervisor: Kenneth Pope