An EEG Study using Coherence

Investigating Response Conflict Processes in 7 and 9Year Old Children: An EEG Study using Coherence Tahani Almabruk*†, Kartik K. Iyer‡, Tele Tan‡, Gareth Roberts*, Mike Anderson* *†Department of Computing, Faculty of Science & Engineering, Curtin University, Perth, Australia ‡ Department of Mechanical Engineering, Faculty of Science & Engineering, Curtin University, Perth, Australia * School of Psychology and Exercise Science, Murdoch University, Perth, Australia Email: [email protected]*†, [email protected]‡, [email protected]‡, [email protected]*, [email protected]* Abstract—Assessing cognitive development in children is of paramount importance in understanding the development of critical neural pathways of the brain. In particular, recognition of stimuli, task accuracy and response time are key features that can inform on stages of brain cognition with respect to age and within age group differences. In this study we investigate neurophysiological responses of the Eriksen Flanker task – an experimental paradigm for assessing attention and cognition - in middle childhood ages (seven-nine years). We analyse EEG data in two age groups: 45 healthy subjects aged seven years with a follow-up study on the same subjects at age nine years. We examine spectral coherence – a method for analysing the correlation between electrode pairs - for all possible combination of pairs. Comparisons of coherence values based on Flanker task conditions (incongruent versus congruent) were assessed in each age group. Consequently, these assessments were used as indicators to the cognitive conflict induced by Flanker incongruent stimuli. For both age groups (seven and nine years) inter-hemispherical coherence increased in the right hemisphere. Moreover, the older children showed less Flanker conflict compared with children aged seven years, especially within the theta band. This decrease in the effect of the cognitive conflict may indicate age-related cognitive developments. Keywords—coherence; Flanker conflict; cognitive development

I.

INTRODUCTION

Children’s cognitive abilities and critical thinking skills develop at a rapid rate following infancy and late childhood. The development of a child’s cognitive abilities was first characterised by the psychologist Jean Piaget, who posited that cognitive development consisted of four stages [1]. The sensorimotor stage was defined from birth to two years, preoperational stage from two to six years, concrete operations stage from six to 12 years and formal operations stage from 12 to 18 years. Our datasets (seven-nine years) fall in the age range of the concrete stage where children form critical abilities to understand logic and progressively improve in conceptualizing concrete issues. This stage of cognitive development has been targeted by several studies for extensive investigations and as a result, behavioural and cognitive differences were found between children aged seven and nine years ( e.g., [2] [3], [4]) . Pozuelos et al. (2014) conducted a study on a group of children aged six to 12 years to investigate the cognitive developments that a year makes. They used a

modified version of a model was introduced by Fan et al. (2002) to assess Attention Networks Task (ANT) [5]. This model was built on Flanker task where the target stimulus is flanked by distracting information. In the case of congruent, these distractions point to the direction of the target (no conflict is induced) whereas in the case of incongruent, they point to the opposite direction (conflict is induced). The induced conflict by Flanker incongruent stimuli was scored for each age group as follows: for each task condition, 1) median reaction time (RT) was taken per subject. 2) The mean was performed over the medians for each condition. 3) Thereafter, conflict score was measured by subtracting the congruent mean from the incongruent mean. This score was used to estimate the efficiency of selective attention network – i.e. attention to a particular part of a stimulus. Their finding which is of interest to us, stated that a significant decline in the effect of Flanker conflict was observed after age seven [2]. The current paper investigates the effect of Flanker conflict on the performance of 45 children aged seven years and its effect on the same dataset two years later. Moreover, it investigates age-related changes in the cognitive conflict. We propose using 1) EEG coherence measurements between electrode pairs instead of the RTs and 2) a cumulative procedure over EEG coherence differences (incongruent minus congruent) instead of subtracting the means of the medians RTs. Different regions of the brain are activated in response to visual task stimuli. The EEG provides a rapid acquisition of spatial and temporal neurophysiological activity through electrodes placed on the scalp [6]. Several methods have been used to investigate the correlation between brain regions and the neurophysiological activities within EEG, such as spectral coherence and phase synchronization [7]. These methods are classified as either linear or nonlinear depending on the underlying hypothesis used to represent the transmission media. For example, if a study is built on the hypothesis that the transmitted pathways are linear then linear methods such as spectral coherence can be deployed. The occurrence of coherence increase in a particular electrode pair across the majority of subjects under Flanker incongruent condition (conflict is induced) compared to Flanker congruent condition (no conflict is induced) could be used as an indicator of the cognitive conflict effect. In this study, we test this hypothesis in addition to the expected

decrease in the conflict effect as age increases. To our knowledge, no follow-up study was reported on using EEG coherence to determine Flanker conflict and its age-related changes. Section II will introduce the method, subject participant data, test protocol and data analysis methods. Section III will present our analysis results while the related discussion and conclusions will be provided in section IV. II.

MATERIALS AND METHODS

A. Methods This research made use of an existing data set that was provided by the Neurocognitive Development Unit (NDU) at the University of Western Australia under a formal collaboration framework with Curtin University. Written informed consent was requested from children and their parents. In addition, an approval was provided by the ethics committee of the school of Psychology, university of Western Australia. B. Subjects Two data sets of healthy children were used in this study. In the first data set, EEG recordings were collected from a group of children aged 7 years (n=45; 23 female; 22 male; mean=7.5 years; STD= 0.27) while performing Flanker task. Afterward, this task experiment was repeated on the same group of subjects two years later in order to produce a second data set (n=45; 23 female; 22 male; mean=9.56 years; STD= 0.26), allowing for longitudinal analyses. C. Task Procedure As described in [8], a friendly model of Flanker task was introduced to the subjects as a game, where a set of five fish are presented on computer screen. In order to respond to the task, children were asked to concentrate on the direction of the central fish. By manipulating the flanking fish directions, three different trial types were produced: 1. Congruent, the stimulus contains five green fish pointing to the same direction (Fig.1 (a)). 2. Incongruent, the stimulus contains five green fish and the direction of the central fish is opposite to the flank fish (Fig.1 (b)). 3. Reversed, the stimulus contains five red fish pointing to the same direction (Fig.1(c)). In order to ensure that children fully understood how to perform the task, instructions and practice trials for each trial type were administered. D. Data Acquisition Data were collected using an Easy-CapTM and electrode impedance level was kept below 5k Ω. The signals were amplified with a NuAmps 40-channel amplifier and digitized at 250 Hz using a linked-mastoid reference. Thereafter, zero phase shift band-pass filter from 0.05 to 30 Hz was digitally performed [8].

E. Data Analysis Pre-processing started by rejecting eye electrodes VEOGL and VEOGU, and an additional three channels were unconnected during the EEG recordings (CPZ, X7 and X8). An averaged mastoid reference was used for all analyses as suggested in [9], to avoid the erroneous estimates in coherence when considering the reference channel. Consequently, signals from the mastoid electrodes (M1 and M2) were excluded and then the analysed electrodes in this study were reduced to 33. A child’s data was removed if multiple button responses were recorded, in addition to any segments with responses shorter than 500 or longer than 2,000 . Epochs were extracted from -600 to 1000 and baseline corrected between -600 and -400 . Fig. 2(b) shows that for each subject in each age group, two new data subsets of epochs were produced. The first dataset contains stimuli with correct behavioural responses at the congruent condition whereas the second one contains stimuli with correct behavioural responses at the incongruent condition. Since the current analysis compares coherence calculations across two conditions, it has been recommended to use an equal number of epochs across the conditions for each subject [10]. This step is important because differences in the epoch count across experimental conditions may have a significant impact on the measurements and statistical analysis. Several methods were addressed in [10] to match the number of epochs across the conditions. In this study, epochs were matched randomly across the conditions for each subject with a minimum of 25 epochs to ensure better estimations for the coherence [9]. This means for each subject , the task condition with the smallest epoch count was kept with no change while an equal number ( ) of epochs was chosen randomly from the other condition. Spectral coherence measures were derived from 1 to 30 at a frequency resolution of 1 . The calculations were performed on the three extracted frequency bands (theta, alpha and beta). Theta was defined from 4 to 7 , alpha from 8 to 12 and beta from 13 to 30 . Spectral coherence between signals is defined as [11] :

and

at frequency

where is the cross-spectrum of signals and . and are the auto-spectra of signals and respectively. The terms , and were averaged over the non-overlapping epochs ( ) before applied to (1) as indicated by the . Coherence ( ) takes a value in the range of [0 symbol 1]. If it is equal or around zero that indicates no correlation is

Figure 1 Flanker task conditions where the crosshair at the middle of the screen is the fixation point.

Figure 2 (a) Coherence matrix as represented by the upper triangular matrix. (b) Coherence matrix computation flow.

expected between the signals, while if it is equal or around one that indicates a strong correlation. For each frequency band, averaging over frequencies was needed to end up with an individual coherence value for each electrode pair. Thereafter a 33 33 coherence matrix was produced with 1089 possible coherence values. The symmetry feature of coherence matrix and its unity of the diagonal elements reduced the calculated values to be 528 unique coherence values per frequency band (Fig.2 (a)). Multiplied by three (frequency bands) in turn results in a total of 1584 unique coherence values for each task condition per subject. If we consider this large amount of coherence measures was calculated for subject at age seven years, similar calculation was performed for the EEG recording of the same subject at age nine years (Fig. 2(b)). The current analysis can be divided into three main steps: 1. Exploring the effect of Flanker conflict on coherence measures of group aged seven years. 2. Exploring the effect of Flanker conflict on coherence measures of group aged nine years. 3. Identifying age-related cognitive developments by subtracting the obtained results from step (1) and (2) in the level of topography figures. The executed procedures to perform step (1) and (2) were identical. Congruent and incongruent coherence matrices were compared frequency band to frequency band. For example,

congruent coherence matrix within theta for subject was subtracted from incongruent coherence matrix within theta for the same subject to produce a difference matrix ( ) as explained in Fig. 2(b). As a result, 45 difference matrices per frequency band were produced for each age group. The difference matrices for each frequency band were summarized in one cumulative matrix (Fig.3 (a)). The fundamental idea behind this matrix as shown in (2) is based on element indices ( , ) in the difference matrices ( ).

( , ) is the difference matrix of subject at the where ( , ) is the indexed EEG electrodes and while produced cumulative matrix. Each time, elements with identical indices in their difference matrices were involved in the process of signing a value to the corresponding indexed location in the cumulative matrix ( ( , )). This value was specified by counting how many times the checked coherence difference element was greater than a threshold (σ) that calculated for each frequency band. The significance of coherence deviation (σ) is a statistical measure depends on the independent number of samples (number of epochs and/or number of frequencies) as explained in [7]:

Figure 3 (a) The cumulative matrix over coherence increases within theta band across subjects aged 7 years. (b) The selected electrode pairs with the top maximum values (70% of the population or over). (c) The topographical figure corresponding to (b).

where = (number of epochs) * (number of frequencies). One more matrix was produced from the cumulative matrix containing less data information (Fig.3 (b)). All electrode pairs of the cumulative matrix that had values higher than 30 (70% of the population) were picked up for the new matrix. This threshold on the population was the highest percentage we still can see difference in coherences measures across the conditions for most of the frequency bands. Reducing the threshold (< 30) obviously results in more electrode pairs but we were interested in including the highest number of the populations. Finally, the obtained results in Fig.3 (b) were visualized by a topography figure as illustrated in Fig.3 (c). III.

RESULTS

A. Group Aged Seven Years As was explained previously, the topographical figure represents the increase in coherence measurements for each electrode pair at the incongruent condition compared to the congruent across the 45 subjects. It can be seen from Fig.4 (a) that, the increase in coherence measurements within theta band is concentrated in the right hemisphere, where parietal area seems to be a common factor in all occurred increases. In this frequency band, there are increase in the correlations that link midline frontal lobe (midline frontal and midline frontocentral) to the right parietal (FZ/P4, FCZ/P4). In addition to the correlation that links right temporal to right parietal (T8/P4). Fig.4 (b) shows that, the midline frontal and midline frontocentral are both involved again in coherence increase within alpha band in their relationship with the right frontal (FZ/F8, FCZ/F8). In addition to intra-hemispheric coherence increase maps the right frontal area to the left fronto-central (F8/FC1).Within beta band in Fig.4 (c), right fronto-central and right centro-parietal are confronted with coherence increase in their connection with the midline frontal (FC6/FZ, CP6/FZ). Moreover, right fronto-central and right motor are confronted with coherence increase in their connection with the midline fronto-central (FC6/FCZ, C4/FCZ). Increase in the intrahemispherical coherence links left fronto-central with right centro-parietal (FC5/CP6), and left parieto-occipital with right occipital (PO9/O2) are observed as well. B. Group Aged Nine Years Fig. 5(a) shows that, no coherence increase occurs in the group aged nine years at the incongruent condition compared to the congruent within theta band. The case is different within alpha band in Fig. 5(b), this difference is represented by the increase in the correlation between right frontal and right

Figure 5 Topography figures of coherence differences (incongruent minus congruent) in the group aged 9 years within: (a) theta, (b) alpha and (c) beta.

fronto-central (F8/FC2) and between right fronto-temporal and right motor (FT10/C4). The increase in coherence measurements within beta band is depicted in Fig. 5(c). This increase occurs within the right frontal area (FZ/F4) and, between left parietal and occipital protuberance (P7/IZ). IV.

DISCUSSION AND CONCLUSIONS

In this paper, we described how to utilize the cumulative procedure on EEG coherence measurements to investigate the effect of Flanker task conflict on two different age groups. Moreover, we identified the common electrode pairs in each age group that were confronted with increase in coherence regarding the cognitive conflict. In addition to take advantage of the graph theory to visualize these particular electrode pairs of interest. Furthermore, we compared between the resulted topographies in order to investigate age related cognitive developments. The increase in coherence measurements regarding Flanker conflict in the group aged seven years seems to be mostly concentrated in the right hemisphere. This may be explained by Thatcher et al. (1987) findings that coherence measurements of children aged four-six years compared to the youngest were confronted with increase concentrated within the left hemisphere electrode pairs [12]. Thatcher et al. (1987) considered that as an evidence of the thought that left hemisphere preceded the right on the cognitive developments. Therefore, the developments in the left side of the brain may justify the occurrence of no difference between congruent and incongruent coherence measurements in the group aged seven years in the current study. Flanker task conflict seems to have less effect on the same group two years later especially within theta band. The increase in the incongruent coherence compared to the congruent still in the right hemisphere. This is consistent with Thatcher et al. (1987) findings that coherence increments among the age eight-10 years are restricted in the right hemisphere. The decrease in the effect of Flanker conflict on the group aged nine years compared to that aged seven years which was observed in this study is consistent with Pozuelos et al. (2014) findings that the efficiency of the selective attention is increased at age eight to 12 years compared to age seven years. ACKNOWLEDGMENT

Figure 4 Topography figures of coherence differences (incongruent minus congruent) in the group aged 7 years within: (a) theta, (b) alpha and (c) beta.

The authors would like to thank Professor Mike Anderson, School of Psychology and Exercise Science, Murdoch University, Dr Allison Fox, School of Psychology, University of Western Australia and Professor Corinne Reid, School of Psychology and Exercise Science, Murdoch University for the use of the EEG data.

REFERENCES [1] [2]

[3]

[4] [5]

[6] [7]

[8]

[9]

[10] [11] [12]

D. H. Feldman, “Piaget's stages: the unfinished symphony of cognitive development,” New Ideas in Psychology, vol. 22, no. 3, pp. 175-231, 2004. J. P. Pozuelos, P. Paz-Alonso, A. Castillo, L. Fuentes, and M. Rueda, “Development of Attention Networks and Their Interactions in Childhood,” Dev. Psychol., vol. 50, no. 10, pp. 2405-2415, 2014. M. R. Rueda, J. Fan, B. D. McCandliss, J. D. Halparin, D. B. Gruber, L. P. Lercari, and M. I. Posner, “Development of attentional networks in childhood,” Neuropsychologia, vol. 42, no. 8, pp. 1029-1040, //, 2004. M. Anderson, C. Reid, and J. Nelson, “Developmental changes in inspection time: what a difference a year makes,” Intelligence, vol. 29, no. 6, pp. 475-486, 11//, 2001. J. Fan, B. D. McCandliss, T. Sommer, A. Raz, and M. I. Posner, “Testing the efficiency and independence of attentional networks,” Journal of cognitive neuroscience, vol. 14, no. 3, pp. 340-347, 2002. E. Niedermeyer, and F. L. da Silva, Electroencephalography: basic principles, clinical applications, and related fields: Lippincott Williams & Wilkins, 2005. B. Saltzbertg, W. D. Burton Jr, N. R. Burch, J. Fletcher, and R. Michaels, “Electrophysiological measures of regional neural interactive coupling. Linear and non-linear dependence relationships among multiple channel electroencephalographic recordings,” International Journal of Bio-Medical Computing, vol. 18, no. 2, pp. 77-87, 1986. C. Richardson, M. Anderson, C. L. Reid, and A. M. Fox, “Neural indicators of error processing and intraindividual variability in reaction time in 7 and 9 year-olds,” Developmental Psychobiology, vol. 53, no. 3, pp. 256-265, 2011. P. L. Nunez, R. Srinivasan, A. F. Westdorp, R. S. Wijesinghe, D. M. Tucker, R. B. Silberstein, and P. J. Cadusch, “EEG coherency: I: statistics, reference electrode, volume conduction, Laplacians, cortical imaging, and interpretation at multiple scales,” Electroencephalography and Clinical Neurophysiology, vol. 103, no. 5, pp. 499-515, 1997. M. X. Cohen, Analyzing neural time series data: theory and practice, 1 ed., London, England: The MIT Press, 2014. W. van Drongelen, Signal Processing for Neuroscientists : An Introduction to the Analysis of Physiological Signals, Burlington, MA, USA Academic Press 2006. R. W. Thatcher, R. A. Walker, and S. Giudice, “Human cerebral hemispheres develop at different rates and ages,” Science, vol. 236, no. 4805, pp. 1110-1113, 1987.