Elsevier

Pattern Recognition

Volume 45, Issue 6, June 2012, Pages 2123-2136
Pattern Recognition

Brain computer interface control via functional connectivity dynamics

https://doi.org/10.1016/j.patcog.2011.04.034Get rights and content

Abstract

The dynamics of inter-regional communication within the brain during cognitive processing – referred to as functional connectivity – are investigated as a control feature for a brain computer interface.

EMDPL is used to map phase synchronization levels between all channel pair combinations in the EEG. This results in complex networks of channel connectivity at all time–frequency locations. The mean clustering coefficient is then used as a descriptive feature encapsulating information about inter-channel connectivity.

Hidden Markov models are applied to characterize and classify dynamics of the resulting complex networks. Highly accurate levels of classification are achieved when this technique is applied to classify EEG recorded during real and imagined single finger taps. These results are compared to traditional features used in the classification of a finger tap BCI demonstrating that functional connectivity dynamics provide additional information and improved BCI control accuracies.

Highlights

► We develop a method to map long range synchronization patterns in the cortex. ► Functional connectivity is measured via the use of complex network metrics applied to characterize these connectivity maps. ► The temporal dynamics of these connectivity measures are measured via Markov modeling and the method is evaluated on synthetic EEG. ► The method is then applied to achieve very high levels of classification for identifying executed and imagined finger taps. ► These results are compared to ERD/S features traditionally used to allow BCI control via motor imagery. The connectivity method is shown to achieve far higher accuracies across all 15 investigated subjects suggesting that connectivity dynamics are suitable for highly accurate BCI control.

Introduction

A brain computer interface (BCI) is a system which allows control of a computer based on the recording and interpretation of neurological activity. Control is based directly on the activity in the brain and therefore there is no need for a BCI user to make any motor movements in order to operate a computer system. This makes BCIs a potentially very useful tool, allowing individuals who suffer from motor movement impairments the ability to communicate and control some aspects of their environment [1].

BCIs may be used by patients suffering from a range of conditions, from partial to complete paralysis and amyotrophic lateral sclerosis (ALS) [2]. They may also be used by healthy individuals for applications such as music generation [3] or computer game control [4].

One of the more commonly used methods for recording the neurological signals used in BCI control is via the electroencephalogram (EEG), a non-invasive method for recording the summed action potentials of large groups of cortical neurons firing at close to the same time [5]. This method has the advantage of being relatively inexpensive, safe, easy to use and has a very high time resolution allowing EEG based BCIs to respond very quickly to user commands [1].

However BCIs have a poor track record in terms of their reliability and performance. Non-stationarity in the EEG signal, contamination by artifacts – signals of non-cortical origin [6] – and inter-trial/inter-subject variability mean reliably classifying EEG into BCI control conditions is very much an on-going research challenge. There is therefore a need to investigate new and innovative methods for reliably classifying EEG signals.

Phase relationships identified between the recordings of electrophysiological activity generated within different cortical regions may provide information about functional relationships between those cortical regions [7]. In particular phase synchronization between two or more different cortical regions may indicate direct communication, a shared input or a common pathway of information flow between those regions [8]. Distributed cortical regions may thus form phase synchronized networks during – for example; memory retention and cognitive processing tasks [8] – which are reflective of the underlying functional relationships between distinct cortical regions; both neighboring and at distance to one another.

Such relationships between cortical regions – measured by their phase synchronization reflected in the EEG – are used in the control of a BCI in only a few studies [9], [10]. This paper introduces a novel method for analyzing the functional relationships within the EEG based on empirical mode decomposition (EMD) combined with complex network metrics. This method is first evaluated on synthetic data and then applied to analyze the functional connectivity in the EEG during a single finger tap BCI control task.

The temporal dynamics of connectivity are investigated as a means of classification. Therefore Markov modeling of the temporal dynamics of changes in functional connectivity patterns during performance of single finger taps is investigated as a method for classification. Hidden Markov models (HMMs) are applied to characterize and classify trials from both synthetic EEG and the single finger tap recordings. The method offers a number of advantages over other work looking at functional connectivity for BCI control such as [10].

Section 2 first introduces the method used to characterize functional connectivity in the EEG. Section 3 then describes the dataset on which the method will be evaluated and the finger tap data the method will be applied to. Section 4 outlines the results achieved before Section 5 discusses these results in the context of their application to BCI control.

Section snippets

Phase synchronization

Functional connectivity between different cortical regions is defined as communication between the regions relating to a specific cognitive task [11]. The aim of this research is to determine if functional connectivity patterns may be used reliably in BCI control.

Functional connectivity may be identified via a number of metrics, two of the most commonly used being coherence and phase synchronization. Phase synchronization is chosen for use in this study as it provides an amplitude free measure

Synthetic EEG

EEG is simulated with a neural mass model (NMM) [23]. The activity of neural populations is modeled via the inhibitory and excitatory interactions of neighboring groups of neurons. The EEG simulated with this model has a realistic power spectrum [23] and is used in a range of studies such as [8], [24]. A schematic of the NMM is illustrated in Fig. 5.

The NMM accepts a Gaussian signal as input to represent the summation of neural activity from neighboring regions. This is added to the activity of

Synthetic EEG

Multivariate (four channel) synthetic EEG was generated to include one of two different randomly generated patterns of synchronization between pairs of channels at random temporal locations. Both patterns of EEG (P1 and P2) contain the same total amount of synchronization but the synchronization occurs in different temporal and spatial locations.

The EMDPL technique was used to characterize the patterns of phase synchronization present in the data.

Discussion

Connectivity measures between different cortical regions indicate a shared pathway of communication or common inputs to those cortical regions. The nature of inter-cortical connectivity thus provides a measure of how different brain regions communicate during cognitive processes and hence may be used to map short and long range pathways of communication and control [7].

Traditional BCI systems use a range of features to allow a user to control an external device such as a computer system or

Conclusions

The observed increase in functional connectivity in this study occurs at a range of frequencies, most notably between 5 and 15 Hz. This work therefore demonstrates that prior to executed movement (and imagined movement) there is an increase in the level of inter-cortical communication at the Mu rhythm and that this may be used successfully as a highly accurate BCI control signal for identifying taps on a single trial basis.

In terms of application to BCI its important to note that BCI systems

Ian Daly holds a M.Eng. in Computer Science from the University of Reading and has been a Ph.D. student in Cybernetics, School of Systems Engineering at Reading University since 2007. His research focuses on Brain Computer Interfaces (BCIs), machine learning, signal processing, bio-signal analysis, meta-heuristic search techniques, computational intelligence and phase synchronization in EEG.

References (41)

  • E.R. Miranda

    Plymouth brain-computer music interfacing project: from EEG audio mixers to composition informed by cognitive neuroscience

    International Journal of Arts and Technology

    (2010)
  • M. Teplan

    Fundamentals of EEG measurement

    Measurement Science Review

    (2002)
  • S.P. Fitzgibbon et al.

    Removal of EEG noise and artifact using blind source separation

    Journal of Clinical Neurophysiology

    (2007)
  • F. Lotte et al.

    A review of classification algorithms for EEG-based brain-computer interfaces

    Journal of Neural Engineering

    (2007)
  • C. Brunner et al.

    Online control of a brain-computer interface using phase synchronization

    IEEE Transactions on Bio-medical Engineering

    (2006)
  • M. Rubinov et al.

    Complex network measures of brain connectivity: uses and interpretations

    Neuroimage

    (2009)
  • T. Wang et al.

    An efficient rhythmic component expression and weighting synthesis strategy for classifying motor imagery EEG in a brain-computer interface

    Journal of Neural Engineering

    (2004)
  • C. Sweeney-Reed et al.

    Empirical mode decomposition of EEG signals for synchronization analysis

  • E. Huang et al.

    The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences

    (1998)
  • R. Andrzejak et al.

    Indications of nonlinear deterministic and finite-dimensional structures in time series of brain electrical activity: dependence on recording region and brain state

    Physical Review E

    (2001)
  • Cited by (84)

    • Neural component analysis: A spatial filter for electroencephalogram analysis

      2021, Journal of Neuroscience Methods
      Citation Excerpt :

      Furthermore, by imposing a statistical independence criteria ICA is at risk of excluding, or misinterpreting, physiologically meaningful sources of neural activity simply because they exhibit activity that is statistically related to other, physiologically distinct sources. As an example of this, movement control typically involves a functional network of distinct brain regions, including, but not limited to, M1 and the supplementary motor area (SMA) (Stefano Filho et al., 2017; Daly et al., 2012). These brain regions are spatially distinct from one another, but, during movement control, exhibit neural activity that is statistically related.

    View all citing articles on Scopus

    Ian Daly holds a M.Eng. in Computer Science from the University of Reading and has been a Ph.D. student in Cybernetics, School of Systems Engineering at Reading University since 2007. His research focuses on Brain Computer Interfaces (BCIs), machine learning, signal processing, bio-signal analysis, meta-heuristic search techniques, computational intelligence and phase synchronization in EEG.

    Slawomir J. Nasuto has been a lecturer (since 2000) and a Reader (since 2007) in Cybernetics, School of Systems Engineering at the University of Reading. His research interests include machine learning, swarm intelligence, computational neuroanatomy and neuroscience, the role of synchronization in cognitive processing, EEG based Brain Computer Interfaces, relationship between structure and function in individual neurons and their networks. He current research also concerns investigating the computational capacity of networks of biological neurons controlling a robot. Nasuto is a member of the EPSRC College and has acted as a reviewer for BBSRC and ESRC.

    Kevin Warwick has received D.Scs. degrees from both Imperial College, London, and the Czech Academy of Sciences, Prague. He has been Professor of Cybernetics at the University of Reading, Reading, U.K., since 1988. He has published over 400 research papers. His research interests lie in machine intelligence, control, and in the integration of biological and technological entities. Dr. Warwick was made an Honorary Member of the Academy of Sciences, St. Petersburg, and in 2000 presented the Royal Institution Christmas Lectures. He is perhaps best known for a series of pioneering experiments in which he received an implant in his median nerves in order to directly connect his nervous system into the internet.

    View full text