Brain Topography

, Volume 23, Issue 2, pp 205–213

Transmission of Brain Activity During Cognitive Task

  • Katarzyna Blinowska
  • Rafal Kus
  • Maciej Kaminski
  • Joanna Janiszewska
Original Paper

DOI: 10.1007/s10548-010-0137-y

Cite this article as:
Blinowska, K., Kus, R., Kaminski, M. et al. Brain Topogr (2010) 23: 205. doi:10.1007/s10548-010-0137-y

Abstract

The transmission of brain activity during constant attention test was estimated by means of the short-time directed transfer function (SDTF). SDTF is an estimator based on a multivariate autoregressive model. It determines the propagation as a function of time and frequency. For nine healthy subjects the transmission of EEG activity was determined for target and non-target conditions corresponding to pressing of a switch in case of appearance of two identical images or withholding the reaction in case of different images. The involvement of prefrontal and frontal cortex manifested by the propagation from these structures was observed, especially in the early stages of the task. For the target condition there was a burst of propagation from C3 after pressing the switch, which can be interpreted as beta rebound upon completion of motor action. In case of non-target condition the propagation from F8 or Fz to C3 was observed, which can be connected with the active inhibition of motor cortex by right inferior frontal cortex or presupplementary motor area.

Keywords

Transmission of information in brain Propagation of EEG activity Short-time directed transfer function Granger causality Cognitive processes Continuous attention test Working memory Active inhibition 

Introduction

The information processing by the brain is connected with the transmission of activity between spatially distributed functionally specialized networks. The spatial characteristic of the process, that is the localization of the areas involved in a particular task, can be estimated with high precision by means of fMRI or PET studies. However, the temporal relations between specialized brain structures, reflected in electrical activity transmissions effected in the short time scale, are beyond the reach of fMRI and PET techniques. In addition, these techniques do not reveal the dynamical functional relations between the investigated structures. The information about propagation of brain activity and relations between different brain structures may be found from EEG signal provided that the appropriate methods will be used. The estimator which allows for the determination of the propagation of EEG activity is the directed transfer function (DTF) (Kaminski and Blinowska 1991). DTF methods have been widely used, e.g. for localization of epileptic foci (Franaszczuk and Bergey 1998), for estimation of EEG propagation in different sleep stages and wakefulness (Kamiński et al. 1997), for determination of transmission between brain structures of an animal during a behavioral test (Korzeniewska et al. 1997), for estimation of cortical connectivity (Astolfi et al. 2005; Babiloni et al. 2005) and many other applications.

Estimation of dynamical transmission of signals in the short-time scale is possible by means of the short-time directed transfer function (SDTF) (Kaminski et al. 2001; Ginter et al. 2001), which is a version of DTF based on ensemble averaging of trials. SDTF is a multivariate method capable of determination of the activity flow as a function of time and frequency. It has been applied successfully e.g. in the investigations concerning motor related tasks (Ginter et al. 2001, 2005; Kus et al. 2006).

In cognitive tasks, especially those connected with the stimulus perception, memorizing, retrieval and decision making, there are several steps in information processing characterized by specific transmission patterns. Usually, the outcomes of cognitive tests are analyzed by means of fMRI or by solving inverse problems. These kinds of approaches allow for identification of the involved structures, however, the dynamical interaction between them is out of the reach of these methods. The novelty of our approach relies on the fact that it allows for the determination of the dynamical patterns of transmissions between the brain structures involved in processing the information. Such transmission is effected in a short time scale and often reciprocal flows are observed. This kind of propagation cannot be identified by methods such as consideration of the phases of coherences—besides, the bivariate methods may lead to erroneous results (Kus et al. 2004). The method used in the present study—a short-time approach which utilizes information from all recordings—is the only one which shows in this case the dynamic evolution of information processing during a task.

Here we shall consider the propagation of electrical activity during the continuous attention test (CAT). CAT is a test introduced by (Tiplady 1992) and widely applied in clinics, since it gives information on possible attention deficits. Topographic features of event related potentials registered during CAT were investigated by means of low-resolution electromagnetic tomography (LORETA) (Basinska-Starzycka and Pasqual-Marqui 2001). The strictly cognitive activity, reflected by inter-condition differences, was found mainly in the prefrontal cortex structures, corresponding with the surface early P3 evoked potential component identified at the midline frontal derivation Fz. In earlier work (Kus et al. 2008), the propagation patterns for trials synchronized in respect to the moment of stimulus occurrence were analyzed. Here we apply another paradigm, namely synchronization of the trials in respect to the start of the movement.

Materials and methods

Material

The CAT test was performed on a group of 16 young right-handed healthy males (age: 20–30 years). The experiment consisted of the presentation of 720 consecutively displayed geometrical patterns, 120 of which were identical to the preceding one (Fig. 1). The patterns were presented in random time intervals varying from 1.5 to 2.5 s. The target condition was defined as any immediately repeated pattern. The subject was instructed to press the button with his right thumb after the perception of the target. An EEG was recorded from 23 electrodes (10–20 system) referenced to linked mastoids. The sampling frequency was 250 Hz. The signals were filtered in the 15–45 Hz frequency band, since the frequencies of interest encompassed the beta and gamma band. For each target condition during the test, two specific time moments can be distinguished: the presentation of the (repeated) picture and motor reaction onset. The experimental records were aligned according to the reaction time.
Fig. 1

The scheme of the CAT test. Upper part: presented sequence of geometrical patterns. Lower part: timing of the experiment

Methods

SDTF is a short-time version of the DTF (Kaminski and Blinowska 1991). DTF and hence SDTF are based on a multivariate autoregressive model (MVAR). The MVAR model assumes that X(t)—a sample of data at a time t—can be expressed as a sum of its p previous values weighted by model coefficients A plus a random value E(t):
$$ {\mathbf{X}}(t) = \sum\limits_{j = 1}^{p} {{\mathbf{A}}(j){\mathbf{X}}(t - j)} + {\mathbf{E}}(t) $$
(1)
The variable p is called the model order. For a k-channel process X(t) and E(t) are vectors of size k and the coefficients A are k × k-sized matrices.
$$ {\mathbf{X}}(t) = (X_{1} (t), \, X_{2} (t), \ldots , \, X_{k} (t))^{\text{T}} $$
(2)

For the determination of the model order, appropriate criteria have been introduced. Here we have used the Akaike criterion (Akaike 1974).

There are many methods for estimating model parameters from the given data available in the literature. The Yule-Walker method relies on estimating the data covariance matrix. The elements of that matrix are estimated for each realization of the process as:
$$ R_{ij} (s) = {\frac{1}{N - s}}\sum\limits_{t = 1}^{N - s} {X_{i} (t)X_{j}^{\text{T}} (t + s)} $$
(3)
where N is the data record length.
The above presented formalism assumes that the stationary time series are analyzed. In our case the response to a stimulus is a phenomenon producing non-stationary time series. To solve this problem, the data segment can be divided into shorter time windows, where we can assume data stationarity. However, when the data segment is short, the statistical properties of the fitted parameters deteriorate and the estimate becomes unacceptable. We may improve the fit using information from all repetitions of the experiment. Assuming that every repetition is a realization of the same stochastic process we may extend the covariance matrix definition (3) to utilize all (NR) realizations (r in parentheses indexes realizations):
$$ R_{ij} (s) = {\frac{1}{{N_{R} }}}\sum\limits_{r = 1}^{{N_{R} }} {{\frac{1}{N - s}}\sum\limits_{t = 1}^{N - s} {X_{i}^{(r)} (t)X_{j}^{{ (r ) {\text{T}}}} (t + s)} } $$
(4)

For ensemble averaging the trials have to be aligned in respect to a chosen time moment, here we have aligned the records in respect to the start of the finger movement. The window length has to be determined in an optimal way, the time resolution is better when a short data window is used, however for short window the statistical properties of estimate are worse. The rule of thumb is to have at least several times (say, 10) more data points than fitted parameters. It implies the practical limitation that a time window cannot be shorter than 10kpNR. Moreover, the result is smoother when overlapping windows are used.

From the correlation matrix defined by (4) model parameters A(j) can be found. Since in case of brain activity different rhythms have a specific information role and since increase of propagation at a specific frequency can be accompanied by decrease of propagation in another frequency (Ginter et al. 2005), the estimation of transmission should be considered in the frequency domain.

Equation 1 can be easily transformed to describe relations in the frequency domain. After rewriting (1) in the form (the sign of A changed)
$$ {\mathbf{E}}(t) = \sum\limits_{j = 0}^{p} {{\mathbf{A}}(j){\mathbf{X}}(t - j)} $$
(5)
the application of Z transform yields
$$ \begin{array}{lll} {\mathbf{E}}(f) &=& {\mathbf{A}}(f){\mathbf{X}}(f) \\ {\mathbf{X}}(f) &=& {\mathbf{A}}^{ - 1} (f){\mathbf{E}}(f) = {\mathbf{H}}(f){\mathbf{E}}(f) \\ {\mathbf{H}}(f) &=& \left( {\sum\limits_{m = 0}^{p} {{\mathbf{A}}(m){\text{exp}} ( - 2\pi {imf}\Updelta t)} } \right)^{ - 1} \\ \end{array} $$
(6)
f denotes frequency
When considering Eq. 6 we see that all the relations between data channels are contained in the transfer matrix H. We may define the directed transfer function (DTF) that describes causal influence of channel j on channel i at frequency f (Kaminski and Blinowska 1991):
$$ \gamma_{ij}^{2} (f) = {\frac{{\left| {H_{ij} (f)} \right|^{2} }}{{\sum\limits_{m = 1}^{k} {\left| {H_{im} (f)} \right|^{2} } }}} $$
(7)

The above equation defines a normalized version of DTF, which takes values from 0 to 1 producing a ratio between the inflow from channel j to channel i to all the inflows to channel i.

The non-normalized DTF is defined as:
$$ \theta_{ij}^{2} (f) = \left| {H_{ij} (f)} \right|^{2} $$
(8)

This quantity is directly proportional to the strength of coupling between the channels i and j (Kaminski et al. 2001). For two channels the DTF corresponds directly to the Granger causality principle. The concept of Granger causality was introduced on the ground of economics (Granger 1969) and it concerned two channels only. DTF is defined for an arbitrary number of channels and it could be considered as an extension of Granger causality to multidimensional systems (Kaminski et al. 2001). It was demonstrated (by Kus et al. 2004; Blinowska et al. 2004) that only a multivariate (not bivariate) approach allows for a correct estimation of directionality of the flows.

SDTF describing the directed transmission strength between each pair of channels is a k × k matrix of time–frequency maps γij(t, f).

The SDTF is a complicated and non-linear function of the data and its statistical properties are difficult to be expressed in an analytical form. Moreover, its value fluctuates due to non-stationarity of the data. For statistical testing of the differences between the flows during the CAT test performance and in the reference period, we used the approach introduced in Korzeniewska et al. (2008), which we will briefly present here. Let us assume that an SDTF value can be expressed as a smooth trend and noise component: y(t, f) = g(t, f) + e(t, f). During the procedure we substitute the SDTF function y(t, f) for each pair of channels by a simplified surface g(t, f), smoothing the variability of the original function while preserving the general trend of changes of the relation. The unknown functions g(t, f) were estimated using the thin-plate spline model (Ruppert et al. 2003), resulting in the estimates ĝ(t, f).

Each repetition of the measurement was divided in two parts: a pre-stimulus and a post-stimulus period. The pre-stimulus part was treated as a baseline and the question was if the data after the stimulus differ from the baseline value. In practice the null hypothesis of the test was the question if a transmission value in the post-stimulus part g(tPOST, f) is not significantly different from any of the transmission values g(tj, f) in the pre-stimulus period (tj = t1,PRE, t2,PRE,…, tmax,PRE). The g values were considered different if for given tPOST and f they are different for every tj from the pre-stimulus period for the frequency f (separate comparisons with every pre-stimulus time point were made due to non-stationarity of the data). To test such a hypothesis we may consider a difference between the pre- and post-stimulus estimated spline values û(tPRE, tPOST, f) = ĝ(tPRE, f) − ĝ(tPOST, f) and hypothetical SDTF trend values u(tPRE, tPOST, f) = g(tPRE, f) − g(tPOST, f). Such differences will be compared against zero value. Let σ(tPRE, f) and σ(tPOST, f) be standard errors of the ĝ(tPRE, f) and ĝ(tPOST, f) estimators. Applying the law of large numbers we may assume that the distribution of the normalized difference tends to normal distribution:
$$ {\frac{{\hat{u}(t_{\text{PRE}} ,t_{\text{POST}} ,f) - u(t_{\text{PRE}} ,t_{\text{POST}} ,f)}}{{\sqrt {\sigma^{2} (t_{\text{PRE}} ,f) + \sigma^{2} (t_{\text{POST}} ,f)} }}} \sim N(0,1) $$
(9)
Based on that assumption the t-test can be used to compare the estimated transmissions difference û(tPRE, tPOST, f) against a zero value. The corridor of confidence can be calculated for that difference at a given significance level as
$$ \hat{g}(t_{\text{PRE}} ,f) - \hat{g}(t_{\text{POST}} ,f) \pm \alpha \sqrt {\sigma^{2} (t_{\text{PRE}} ,f) + \sigma^{2} (t_{\text{POST}} ,f)} $$
(10)

If such a corridor contains zero value within its range (for at least one pair tPRE and tPOST for the frequency f), the hypothesis about lack of change in transmission holds. The α coefficient is a quantile of the distribution corrected for multiple comparisons, corresponding to the chosen significance level. It is a conservative approach, so we may miss some flows, but rather we should not observe non-significant connections.

Results

In the first step of the analysis the artifacts were thoroughly eliminated by visual inspection. In some subjects the performance of the test was accompanied by eye blinking or muscle artifacts. The data epochs with artifacts were eliminated; for subjects with high occurrence of artifacts whole data records were eliminated. After the procedure 9 out of 16 subjects were left for further analysis.

Since the number of parameters of the model has to be higher than the number of data points, and the number of parameters increases as a square of the number of channels, we had to limit the number of signals simultaneously processed to twelve. We have chosen the electrodes overlying the areas that were found to be involved in the processing of the CAT (Basinska-Starzycka and Pasqual-Marqui 2001).

The trials were synchronized in respect to the motor reaction. Records of 1.5 s length were analyzed. The first 0.5 s before pattern presentation were used as a reference period to assess the significant changes in the brain activity during the test. In order to find a time course of the SDTF, we applied a sliding window of 160 ms length which was moved by two points (8 ms).

We have considered two situations, both corresponding to correct responses: target—pressing the switch when the shown pattern was equal to the previous one, and non-target—(true negatives) not pressing the switch when the pattern was different from the previous one. The number of false reactions was too small to statistically estimate the SDTF functions by ensemble averaging.

An example of significant changes in flows of EEG activity in respect to the reference period assessed according to the statistical procedure described in “Materials and Methods” section, is shown in Fig. 2 for target and non-target conditions. In the picture, the significant transmission from the electrode marked above the column to the electrode marked at the left of the picture is shown as a function of time and frequency. By inspecting the plot, the general impression is that there is more propagation from the frontal electrodes (at least in higher frequencies) than from the central and posterior electrodes. However, plots of this kind are difficult to inspect.
Fig. 2

Significant (in respect to reference period) changes of flows obtained for target (upper part) and non-target (bottom part) conditions. In each small box are shown: the change of propagation from the electrode marked at the top of the picture to the electrode marked at the left. Horizontal axis: time in seconds, vertical axis: frequency in hertz. Shades of yellow to red correspond to increases and shades of blue to decreases of flows. Black vertical lines mark the presentation of the CAT pattern and the motor reaction

In order to visualize the patterns of transmissions better, we have integrated the significant flows in the frequency band (25–45 Hz) and we have constructed movies representing the dynamically changing propagation patterns for all patients, for target and non-target conditions. We have concentrated on high beta and gamma band, since they play a role in the processes connected with attention and decision making (Bekisz and Wróbel 1999). The animations presenting the dynamical propagation of EEG activity are accessible on the Internet at the address: http://brain.fuw.edu.pl/~kjbli/CAT_MOV.html.

It is difficult to establish a universal pattern of the dynamics of activity propagation, since there was a large inter-subject variability connected with the different reaction times and different strategies in solving the task, however some general trends were observed. In Figs. 3, 4, and 5 snapshots from movies of three subjects for target and non-target conditions are shown. The pictures were chosen in a way to show the characteristic flows present in most of the subjects and appearing during several tens of milliseconds, in a sustained way. Nevertheless some arrows on the pictures may show transient flows. Three different epochs are distinguished. The timing of the snapshots is different, since subjects had different reaction times. In the first epoch, which can be connected with the mental comparison of displayed patterns and decision making, the propagation from the prefrontal electrodes and between prefrontal and frontal electrodes was common for all subjects for both conditions. The second epoch was rather a “silent” period in case of the target condition: not much transmission between electrodes was observed. At the end of the task, after the switch pressing, propagation from C3 toward frontal electrodes was visible for all subjects. This kind of transmission was only occasionally present for the non-target condition.
Fig. 3

Snapshots from the movie presenting significant changes in transmissions in one subject, for target (upper) and non-target (lower part). Intensity of flow changes for increase: from pale yellow to red, for decrease: from light to darkblue. In the right upper corner the time after cue presentation (in seconds)

Fig. 4

Snapshots from the movie presenting significant changes in transmissions for one subject for target (upper) and non-target (lower part). Intensity of flow changes for increase: from pale yellow to red, for decrease: from light to darkblue. In the right upper corner the time after cue presentation (in seconds)

Fig. 5

Snapshots from the movie presenting significant changes in transmissions for one subject for target (upper) and non-target (lower part). The intensities of flow changes: increase—from pale yellow to red, decrease—from light to darkblue. In the right upper corner the time after cue presentation (in seconds)

In case of non-target, the transmission from the frontal electrodes and between them was usually present in the first epoch. Then in most of the subjects (6 out of 9) propagation from F8 to C3 was observed (in case of two subjects it was present also for the target condition, but at a later stage of the task). In one of the subjects, the propagation from F8 to C3 was rather weak, but the transmission from Fz to C3 was remarkable. In cases when there was no propagation from F8 to C3, propagation from Fz to C3 was present. This kind of propagation pattern is shown in Fig. 5.

Blue arrows in Figs. 3, 4, and 5 illustrate significant decrease in propagation in respect to the reference period. Usually these decreases were visible in posterior and central electrodes, in particular Cz. We can conclude that these locations were not involved in the information processing at a given moment during the task.

Discussion

The patterns of transmissions described above correspond well with the neurophysiological evidence reported in earlier studies. In both conditions of the CAT test, especially in the first phase, we have observed propagation from the prefrontal and frontal structures, which are considered to be active in the tasks involving pattern memorization, retrieval, comparison and decision making. The frontal lobes and especially the prefrontal cortex (PFC) were reported to be responsible for information storage and working memory (Romo et al. 1999). Smith and Jonides (1999) concur (in agreement with the majority of researchers) that executive processes are mediated by the PFC, which is involved in the regulation of processes operating on the contents of working memory. They include focusing attention and updating and checking the contents of working memory to determine the next step in a sequential task. The evidence of the role of frontal cortex in the tasks involving working memory is well founded, since it comes from studies of patients with excisions of the frontal cortex (Owen et al. 1990), from electrophysiological recording works in animals (review in Goldman-Rakic 1987), from PET (Owen et al. 1996) and fMRI (Zarahn et al. 2006) experiments. It was also reported that PFC is involved not only in working memory tasks, but also in encoding, retrieval, manipulation and monitoring (Zarahn et al. 2006) and that there is a reciprocal flow of information between the frontal and parietal cortex (Quintana et al. 1989; Chafee and Goldman-Rakic 2000). Indeed we have observed transmissions between the frontal and parietal areas (Figs. 3, 4, and 5).

In almost all subjects, in the target condition after the switch pressing there was propagation found from the C3 electrode, which overlays the hand motor area. This observation is compatible with a well known phenomenon connected with hand movement (Pfurtscheller and Lopes da Silva 1999; Kus et al. 2006).

An interesting phenomenon is the transmission from the F8 derivation located over the right inferior cortex (rIFC) to C3. It can be connected with the active inhibition. In several imaging studies, rIFC was identified as the site responsible for the inhibition in “go/no go” tasks (Aron et al. 2003; Forstmann et al. 2008). rIFC was found to play an inhibitory role across a range of tasks requiring suppression of response tendencies. Forstmann et al. (2008) reported that individuals revealing strong rIFC activation in implementing inhibitory control also exhibited higher white matter coherency at an adjacent anatomical site. This observation points to the possible easy transmission from rIFC.

In some subjects in our study the propagation in non-target condition was observed from Fz instead of F8. Fz is localized over the presupplementary motor area (preSMA). Robust preSMA activation was found by Aron and Poldrack (2006) in an fMRI study on standard stopping. Microstimulation studies in human subjects have identified an inhibitory motor region in the preSMA where stimulation produces hesitation or arrest of manual and vocal responses (Luders et al. 1988; Fried et al. 1991). Burle et al. (2004) suggested the inhibitory role of preSMA as well. In the context of the CAT experiment it is interesting that, as it was pointed out by Aron et al. (2007), preSMA plays also a conflict detection/resolution role. The fact that in some subjects the active inhibition was connected with activity spreading from rIFC in others from preSMA (or both) may be explained by the different strategies of task solving by the subjects. The transmission between rIFC and SMA observed by us is in agreement with the evidence reported by Aron et al. (2007) about existence of a tract between preSMA and IFC.

In our study, we have observed quite often transmissions between rather distant locations like F8 and C3. This finding is not surprising, since it was reported by Burle et al. 2004 that inhibition of motor structures originates from long-range cortico-cortical connections.

We have also observed decreases of propagation in respect to reference period. These decreases in areas not crucial for the task may be explained as a “shift of resources” away from ongoing, but not essential processes to an increasingly demanding cognitive task (Habeck et al. 2005).

The method of estimation of EEG propagation based on SDTF gives information on processes acting in the scale of tens of milliseconds and allows for understanding the dynamical interactions between the brain structures. Although this information is not so precise in spatial dimension as the one obtained by the BOLD technique, it reflects the brain activity directly and makes it possible to establish the relations between the main centers that are playing a role in information processing.

The evidence obtained in this study is compatible with the knowledge gathered in earlier studies based on invasive electrophysiological techniques or fMRI. However, it goes beyond the information reported in those studies, since it provides insight in the active brain and dynamic interactions accompanying information processing. SDTF is a technique which possibilities only start to be recognized. In our opinion, it might open interesting new perspectives in research on cognitive functions.

Acknowledgments

We would like to thank Dr. B. Burle for helpful discussions. This work was supported partly by the COST Action BM0601 “NeuroMath” and grant of Polish Ministry of Science and Higher Education (Decision No. 119/N-COST/2008/0).

Copyright information

© Springer Science+Business Media, LLC 2010

Authors and Affiliations

  • Katarzyna Blinowska
    • 1
  • Rafal Kus
    • 1
  • Maciej Kaminski
    • 1
  • Joanna Janiszewska
    • 1
  1. 1.Department of Biomedical PhysicsUniversity of WarsawWarsawPoland

Personalised recommendations