Elsevier

Neurocomputing

Volume 170, 25 December 2015, Pages 15-31
Neurocomputing

Minimum connected component – A novel approach to detection of cognitive load induced changes in functional brain networks

https://doi.org/10.1016/j.neucom.2015.03.092Get rights and content

Abstract

Recent advances in computational neuroscience have enabled trans-disciplinary researchers to address challenging tasks such as the identification and characterization of cognitive function in the brain. The application of graph theory has contributed to the modelling and understanding the brain dynamics. This paper presents a new approach based on a special graph theoretic concept called minimum connected component (MCC) to detect cognitive load induced changes in functional brain networks using EEG data. The results presented in this paper clearly demonstrate that the MCC based analysis of the functional brain networks derived from multi-channel EEG data is able to detect and quantify changes across the scalp in response to specific cognitive tasks. The MCC, due to its sensitivity to cognitive load, has the potential to be used as a tool not only to measure cognitive activity quantitatively, but also to detect cognitive impairment.

Introduction

The evolution of science and technology in the past two decades has been such that an enormous amount of data are continuously being generated and made available for analysis [1]. The emergence of extremely large, complex patterns in the form of sequences, trees, and graphs in many scientific and commercial applications created a need for a powerful data representation of the entities, their attributes and their relationships to other entities. Graph is one of the most sophisticated data structures useful for modelling, describing and mining complex structures such as internet, web, communication networks, social networks, metabolic networks, and biological networks where the relationships between the objects in the system play a dominant role [2], [3], [4]. Investigation to understand the intricacies of the underlying structural and functional behaviours of complex and dynamic network systems using graph theory has become crucial in various scientific disciplines such as social sciences, systems biology and most recently cognitive neuroscience [5], [6], [7].

Given two complex systems, some sort of numeric rating scale is essential to differentiate their relative complexities. Some of the indicators to predict different levels of complexity are human observation and subjective rating, number of distinct elements, number of parameters controlling the system, minimal description, and information content [8], [9]. The human brain is one of the most complex large-scale adaptive networks ever known. Being a very complex system, estimating the complexity of functional brain networks during various states of functioning poses major challenges due to the non-stationary, non-linear, and time-varying nature of the underpinning neuronal activity [10], [11]. The application of network science has, however, significantly enhanced the understanding, modelling and characterization of complex functional brain networks which has gained traction among the cognitive neuro-engineering research communities in the recent past [12], [13]. Advances in neurophysiological recording of brain activity have provided new investigative avenues to support research on acquiring dynamic and non-trivial information relating to patterns of interactions between functional brain networks [14], [15], [16].

Electroencephalography (EEG) is a non-invasive neurophysiological measurement of the electrical activity caused by the firings of billions of neurons in the brain and is recorded by multi-channel electrodes placed on the scalp [17]. Coordinated electrical activity in different brain regions indicates functional relationships between these regions. Sophisticated data search capabilities, statistical techniques, complex network metrics and graph mining algorithms are thus needed to unfold and discover hidden patterns and associated correlations in the functional brain networks [18]. The associated task is further complicated by complex inter-neuronal processes which combine to create a global expression of cognitive load independent of the contributions of individual processes. Thus, the essential challenge to the research community is to identify the subtle changes in various regions of brain during different states of brain activity.

Many past and recent researchers have focused on the abnormality of brain functioning and distinguished these from the normal brain functioning for clinical applications. Analysis of the normal brain functioning during different brain activities to identify cognition and associated changes with various brain regions has gained attention. Moreover, mental health promotion is one of the major research challenges for governments, universities, research organizations and healthcare industries. Systematic investigation of measures and metrics is essential in the neurological analysis of mental disorders to make reliable inferences and detect the minute changes in different regions of the brain during cognition.

This research work, in general, makes a useful contribution in advancing graph theoretic approach to include the value of a special spanning subgraph to detect more influential connection patterns in a network. In particular, this paper proposes a new graph theoretic concept called minimum connected component (MCC) to detect the underlying neuronal oscillatory patterns of two different states of brain namely the idle state and the cognitive load state as a means of identifying the cognitive activity. It proposes an algorithm that uses a special graph operator to identify and measure the cognitive load induced changes in the functional brain network. The new approach appears to be sensitive to cognitive load induced EEG changes which are otherwise difficult to detect. The MCC, due to its sensitivity to changes in the functional connectivity during cognitive load, has the potential to be used as a tool to not only measure cognitive activity quantitatively, but also detect cognitive impairments and hence may help address mental health issues.

The rest of the paper is organized as follows. A survey of current approaches to identify cognitive activity, the applications of graph and information theoretic approaches to network construction and functional brain network analysis are discussed in Section 2. A detailed description on the proposed methodology of functional brain network analysis for change detection using MCC to identify cognition induced changes is proposed in Section 3. The experimental setup including the data collection and pre-processing is presented in Section 4. Experimental analysis and inferential statistical methods to validate the proposed methodology are discussed in Section 5. We conclude with a summary discussion including future directions for complex functional brain network analysis in Section 6.

Section snippets

Current approaches to identify cognitive activity

Cognition is the result of dynamic interactions between dispersed brain regions resulting from transformation of sensed information into action. As a result, cognition can be considered as coordinated brain activity emerging from the interaction and integration of building blocks such as attention, memory, language, learning, reasoning, problem solving, and decision making [19], [20]. Cognitive neuro-engineering is a reverse-engineering process that deals with understanding human cognition

Functional brain network analysis using minimum connected component (MCC)

Identifying hidden dynamic patterns contained in the underlying neurophysiological signals has posed an essential challenge to model the data as complex networks. Various studies on analyzing the behavioural aspects of functional brain networks have used graph-theoretic and information-theoretic concepts from the perspective of dealing with non-stationary EEG data. Taken together, the correlation, coherence and synchrony of EEG are likely to change (increase or decrease) given a specific

Data collection and pre-processing

Ten healthy adults (9 male and 1 female) volunteered to participate in the EEG data collection. Following experimental briefing, participants were seated in a Simuride driving simulator and given time to practice driving in this virtual environment. Participants were also asked to drive using US standards (right-hand side of the road). Automatic transmission mode was chosen to avoid movement artifacts associated with gear changing while the virtual road chosen contained both straight and

Experimental results and discussion

To illustrate the efficacy of the MCC algorithm, an extensive experimental analysis has been conducted. Firstly, the novelty of the MCC algorithm has been validated on a simple real-time dataset. A set of 25 students studying 6th semester MSc Theoretical Computer Science course were chosen as nodes and the average number of Whatsapp messages exchanged between all the pairs of them as edges for a period of one week. This network is named as baseline (WAppBase). The students underwent internship

Conclusion

The efficacy of MCC in detecting cognitive load induced changes is clearly demonstrated in this paper. The MCC graphs obtained for different brain states are considered as a useful means of thresholding the original complete graphs obtained using NMI. The concept of MCC can further be extended to finding predominant patterns such as induced subgraphs, maximum common subgraphs and cliques by using efficient subgraph mining algorithms to understand the underlying neuronal connectivity and

Acknowledgement

This work is being supported by the Cognitive Neuro-Engineering laboratory (CNeL) Grant No: (DSTO- UniSA) – 2011/1167673/1, University of South Australia.

R. Vijayalakshmi is an Associate Professor at the Department of Applied Mathematics and Computational Sciences, PSG College of Technology. She has over 18 years of teaching and research experience. She has received PhD in Computer Science (Thesis title: Efficient Algorithms and Data Structures for Storage and Mining in Large Graph Databases) from Anna University, Chennai, India in 2013, MCA (Master of Computer Applications) from Bharathiar University, Coimbatore, India in 1996. Under the

References (62)

  • G. Nolte et al.

    Identifying true brain interaction from EEG data using the imaginary part of coherency

    Clin. Neurophysiol.

    (2004)
  • A. Zalesky et al.

    Network-based statisticidentifying differences in brain networks

    NeuroImage

    (2010)
  • E. van Dellen et al.

    Epilepsy surgery outcome and functional network alterations in longitudinal mega minimum spanning tree analysis

    NeuroImage

    (2014)
  • A. Delorme et al.

    EEGLAB: an open source toolbox for the analysis of single-trial EEG dynamics including independent component analysis

    J. Neurosci. Methods

    (2004)
  • J. Han et al.

    Data Mining: Concepts and Techniques

    (2006)
  • M. Deshpande, M. Kuramochi, G. Karypis, Automated approaches for classifying structures, in: Workshop on Data Mining in...
  • L.C. Freeman

    The Development of Social Network Analysis: A Study in the Sociology of Science

    (2004)
  • S.H. Strogatz

    Exploring complex networks

    Nature

    (2001)
  • E. Bullmore et al.

    Complex brain networksgraph theoretical analysis of structural and functional systems

    Nat. Rev. Neurosci.

    (2009)
  • L. Sean et al.

    Analyzing complex functional brain networksfusing statistics and network science to understand the brain

    Statist. Surv.

    (2013)
  • Yaneer Bar-Yam

    Dynamics of Complex Systems (Studies in Nonlinearity)

    (1997)
  • T.M. Cover et al.

    Elements of Information Theory

    (1991)
  • G. Dumermuth

    Quantification and analysis of the EEG

    Schweiz. Arch. Neurol. Neurochir. Psychiatr.

    (1974)
  • S. Herculano Houzel et al.

    Isotropic fractionatora simple, rapid method for the quantification of total cell and neuron numbers in the brain

    J. Neurosci.

    (2005)
  • O Sporns et al.

    The human connectome: a structural description of the human brain

    PLoS Comput. Biol.

    (2005)
  • P.L. Nunez

    Electroencephalography

    Encycl. Hum. Brain

    (2002)
  • V. Eckardt Barbara

    What is Cognitive Science?

    (1995)
  • H.E. Atallah et al.

    Hippocampus, cortex, and basal gangliainsights from computational models of complementary learning systems

    Neurobiol. Learn. Mem.

    (2004)
  • M. Malaak et al.

    Changes in cognitive state alter human functional brain networks

    Front. Hum. Neurosci

    (2011)
  • Y. He et al.

    Graph theoretical modeling of brain connectivity

    Curr. Opin. Neurol.

    (2010)
  • M.E.J. Newmann

    The structure and function of complex networks

    SIAM Rev.

    (2003)
  • Cited by (0)

    R. Vijayalakshmi is an Associate Professor at the Department of Applied Mathematics and Computational Sciences, PSG College of Technology. She has over 18 years of teaching and research experience. She has received PhD in Computer Science (Thesis title: Efficient Algorithms and Data Structures for Storage and Mining in Large Graph Databases) from Anna University, Chennai, India in 2013, MCA (Master of Computer Applications) from Bharathiar University, Coimbatore, India in 1996. Under the auspices of the MoU signed between PSG College of Technology and University of South Australia, she has established the Computational Neuroscience Lab (CNsL) at PSG College of Technology in November 2012. She is currently associated in the collaborative research project entitled ‘Cognitive Modeling and Analysis Research’, sponsored by Professor Nandagopal D., Dean: International, Division of Information Technology, Engineering and the Environment, University of South Australia (UniSA), supported by Cognitive Neuro-Engineering Lab (CNeL) at UniSA. She is a Visiting Academic at UniSA and currently an Associate Supervisor for a full-time PhD Scholar at UniSA.

    D. Nandagopal brings to bear a unique combination of leadership and management in Strategic Science, as well as experience across a broad range of disciplines. His career in research spans three continents (Asia [India], North America and Australia) covering a wide range of areas in Science and Engineering including Defence Science, Electronics, Biomedical Engineering and Sensor Signal and Information Processing. He has significant experience at systems level, especially in high level architectures, mission and combat systems, autonomous systems. He has developed a passion for Systems and the system of Systems Research.

    Professor Nandagopal joined UniSA after extensive experience with the Australian Department of Defence. He has been with the Defence Science and Technology Organisation (DSTO) for over 23 years, during which time he held various Senior Executive positions including the position of Deputy Chief Defence Scientist. He has also served on various Defence Committees in Canberra. Professor Nandagopal has taken up the position of Dean in the International and Chair Defence Systems at UniSA after leaving DSTO in August 2012.

    Professor Nandagopal has held academic positions at the University of Adelaide, McMaster University (Canada) and the University of Melbourne. He has also held Adjunct positions at the University of Adelaide and the Australian National University (ANU). Professor Nandagopal currently leads research in Systems, which includes the following topics: 1. intelligent integrated guidance, navigation and control; 2. passive vision based navigation of UAVs; 3. service oriented architectures for naval combat systems; 4. cognitive modelling; and 5. identification of cognitive activity and functional brain networks by computer visualization and mapping of EEG data. Professor Nandagopal has also established a Cognitive Neuroengineering Laboratory (CNeL) and an Intelligent systems laboratory to assist the six PhD students he currently supervises.

    N. Dasari is a PhD Research Scholar at the University of South Australia Research abstract Electroencephalography (EEG) is the recording of the electrical activity of the brain from the scalp. EEG in general reflects brain functioning and is used generally to diagnose any neurological disorders. Extensive EEG studies have been conducted to characterize various neuro-physiological and mental states. The proposed PhD study aims to identify cognitive activity from EEG and hence characterize cognitive functions. Research study will investigate advanced computer visualization techniques to map the neural activity relating to cognitive states and hence assist the development of cognition enhancing methodologies. The outcomes of this study will enable better understanding of cognitive functions and may contribute towards addressing mental health issues.

    B. Cocks is a Research Assistant at the University of South Australia. Her qualifications are as follows: Doctorate in Psychology, in progress; 2012 Bachelor (Honours) Psychology, completed in 2008, University of New England; and Advanced Diploma Arts, completed in 2007. Ms. Cocks returned to full time study in 2005 after working in media and advertizing. She graduated with a Bachelor of Psychology with First Class Honours in 2008 from the University of New England, Armidale, NSW. Her thesis title ‘The Grammatical Class Files: The Lexical Stress Is In There’ examined the effects lexical stress and grammatical class on word-level language processing at a neural level using EEG. She also received the Keith and Dorothy Mackay Honours Scholarship for the year.

    Ms. Cocks is currently, at the business end of a PhD in neuropsycholinguistics looking at the neural substrates of grammatical class perception (nouns vs. verbs) with an emphasis on mirror neuron involvement. She is also looking at whether mirror neurons are involved with language learning in adults and, if so, whether that involvement is differential according to grammatical class type.

    Ms. Cocks is interested in addicitive behaviours, language perception and developmental disorders such as autism spectrum disorders (ASDs). She has also recently become interested in post-traumatic stress disorder (PTSD) specifically as it relates to defence force personnel. Ms. Cocks has tutored/marked/mentored a variety of subjects in psychology and medicine, ranging from first year public health to third year human neuroscience.

    N. Dahal is a PhD Research Scholar at the University of South Australia Research abstract The cognitive ability of combat personnel vary greatly due to injuries, battle stress, fatigue, attention lapses and other distractions. As a result combat personnel returning home from deployment experience a number of mental health conditions such as post-traumatic stress disorder (PTSD) and cognitive impairments. There have been attempts to understand and enhance cognition qualitatively using both medical and psychological interventions. However, there is little effort in developing metrics for quantifying and hence understanding cognition. The proposed research study therefore aims to investigate techniques and methodologies to characterize cognition (develop metrics) using advanced signal processing and cognitive modelling techniques.

    M. Thilaga is working as an Assistant Professor in the Department of Applied Mathematics and Computational Sciences, PSG College of Technology since 2008. Her area of interest is complex network analysis. She has been associated with the Computational Neuroscience Lab established at PSG College of Technology in collaboration with University of South Australia since 2012. She is currently pursuing PhD under Anna University, Tamil Nadu, India.

    1

    Computational Neuroscience Laboratory, PSG College of Technology, India.

    2

    Cognitive Neuroengineering Laboratory, University of South Australia, Australia.

    View full text