EEG SIGNALS IN EPILEPSY AND MIGRAINE

Analysis and Simulations by Multi-agent Systems

Alessandro Vigan

o, Neri Accornero

C.I.S.B. InterDept. Research Center, Sapienza University - Roma, Italy

Alfredo Colosimo

Dept of Neurology, Sapienza Universita’ - Roma, Italy

Keywords:

EEG, Multi Agent Systems, Epilepsy, Migraine.

Abstract:

The preliminary results of some observations carried out on the spectral content of EEG signals from migran-

ious and epileptic individuals and, in particular, on the spatio-temporal correlation of the neuronal activation

in the two pathologies, are presented. In the aim to simulate the qualitative features of EEG signals associated

to migraine and epilepsy, we used a computational approach based upon Pearson correlations and a Multi

Agent System. Our ﬁndings, although still not conclusive, revealed considerable heuristic power on the sole

assumption of a similar synchronization process of the underlying neuronal population, and may provide in

the long term useful hints to a very difﬁcult problem.

1 INTRODUCTION

According to the World Health Organization, epilepsy

is one of the most common neurological disorder and

its prevalence in the total world population is of the

order of about 1%, with no major geographical bias.

Migraine, on the other hand, although less severe as a

disorder, is much more common and its well known

worldwide distribution, according to some hypothe-

sis, linked to economic levels and life styles (Lipton

and Bigal, 2005).

It has been suggested some time ago (Sack, 1992;

Ottman and Lipton, 1996) that epilepsy and migraine

are correlated, indicating that a crucial role in both

pathologies is played by an abnormal synchronization

of the involved neuronal populations. This has been

recently reassessed (Rogawski, 2008) on the basis of

accurate epidemiological data (Fig. 1). Thus, re-

cruiting a larger and larger number of phase-coupled

neurons, should account for: i) the peculiar activity

bursts appearing in EEG signals; ii) the close tem-

poral correlation of the activity bursts with macro-

scopic clinical symptoms like epileptic seizures or in-

dividual perceptions like visual aura; iii) the typical

rythmic occurrence and spatial patterns of the activ-

ity waves. Such apparently simple phenomena ap-

pear amenable to simulation, taking advantage of the

continuous increase in hardware power and ﬂexibil-

ity/sophistication of simulation environments (Brette

et al., 2007).

We report here the preliminary results of a study

on the common features of EEG signals associated to

migraine and epilepsy which include: i) a systematic

correlation of the spectral content of the EEG signals

recorded from individuals with diagnosis of focal and

diffused epilepsy and of migraine, and ii) a simulation

study of the shift from random to synchronous activity

within an artiﬁcial Multi Agent System.

2 METHODS

2.1 EEG Records and Exclusion

Criteria

The EEG signals analyzed in this work have been

recorded in the Dept. of Neurological Sciences of

Univ of Rome - Sapienza, according to the stan-

dard protocol (Flink et al., 2002) and using a 10–

20 montage, except in the case of the migraine sig-

nals, which came from the Australian EEG Database

(Hunter et al., 2005). The exclusion criteria used by

clinicians to select the signals included the absence of

447

Viganó A., Accornero N. and Colosimo A. (2010).

EEG SIGNALS IN EPILEPSY AND MIGRAINE - Analysis and Simulations by Multi-agent Systems.

In Proceedings of the Third International Conference on Bio-inspired Systems and Signal Processing, pages 447-451

DOI: 10.5220/0002750204470451

 SciTePress

56%

24%

17%

Non Migraine-affected

people

Interictal Migraine

Postictal Migraine

Preictal Migraine

Figure 1: Statistics about the migraine and epilepsy comorbidity. The illustration shows the prevalence of migraine disease in

a Norwegian epilepsy-affected population. (Data from table 2 in (Syvertsen et al., 2007).)

any pharmacological, psychiatric or behavioral inter-

ference potentially able to produce signal alterations.

2.2 Data Analysis

The software toolset used in this work included a

number of macros written in the programming lan-

guage of MatLab (Mathworks, 2006), JMP (SAS,

2007) and NetLogo (Wilensky, 2009), and are freely

available upon request. The data analysis procedure

can be summarized in the following steps:

• The digitized EEG signals produced by the classi-

cal 10 - 20 montage of the electrodes, described

in (Flink et al., 2002), were carefully cleared

from artifacts, as identiﬁed by the clinical experts.

Whenever the artifacts were only present in some

of the signals, all records from that montage were

submitted to the identical clearing procedure, in

order to preserve their phasing.

• The digitized signal (recorded at a 256 Hz) from

each electrode was ’windowed’ in stretches of

about 8 sec. For example, from a 5-minutes-

lasting record, 62 windows were obtained, which

typically reduced to about 40 after the above de-

scribed artifact clearing.

• The spectral content of the signal in each win-

dow was obtained by a macro based upon the DFT

procedure of MatLab (Mathworks, 2006), and the

power spectra of the windows derived from each

electrodic record were aligned in a matrix of typi-

cal size = 20 (windows) * 50 (frequencies in Hz).

• The above matrices (corresponding to whole elec-

trodic records) were correlated by means of the

Pearson correlation coefﬁcient (R) :

R =

∑

i=1

−

Y )(X

−

∑

i=1

−

Y )

∑

i=1

−

(1)

Positive and negative values indicate that the two

variables show the same or, respectively, an oppo-

site trend with respect to each other. Close to zero

values indicate the absence of any signiﬁcant link-

age. For a complete survey of the Pearson Corre-

lation Coefﬁcient as a powerful data analysis tool,

see (Rodgers and Nicewander, 1988).

2.3 Multi Agent System (MAS)

Multi agent systems (MAS) are useful for simulating

the highly cooperative behaviour of individuals in so-

cial groups like human communities, insect colonies

nests or multicellular organisms (Russell and Norvig,

2005). In a neuroscience context, the architecture of

the agent system is such that each agent corresponds

to a neuron or a neuron class and is able to send

signals according to its neighbours, thus inﬂuencing

their activation state. The activation time of the sin-

gle agent may also change according to afferences of

connected ﬁbers, regulating the activation threshold

as well as the ﬁring frequency. A MAS system should

be able to reproduce the EEG signal typical of mi-

graine or epilepsy. The more or less realistic con-

ditions under which this may be achieved, could be

taken as reinforcing/disproving the theory that the ba-

sis of both pathologies is a common synchronization

mechanism.

3 RESULTS

3.1 Correlating Signals from Homo -

and Controlateral Electrodes

An interesting trend is shown by correlating traces

from symmetric couples of electrodes in the two

BIOSIGNALS 2010 - International Conference on Bio-inspired Systems and Signal Processing

448

hemispheres within the same subject. Table I con-

tains the Pearson coefﬁcients obtained by coupling: i)

corresponding electrodes in the two hemispheres, or-

dered in the rostro-caudal direction, from the fronto-

parietal electrodes (F

) to the occipital (O

)

lobes (columns 1,2,3), and ii) electrodes of the same

(left) hemisphere (columns 4,5).

Table 1: Time dependent Pearson correlations between left

and right emisphere in different pathologies ( S1, S2 = Mi-

graine; S3 = diffused epilepsy; S4 = focal epilepsy; S5 =

control). The correlations were calculated from the records

of the following couples of electrodes: Fp = frontal-pole; T

= temporal; O = occipital. Odd and even sufﬁxes refer to

right and left emispheres, respectively; values higher than

0.66 are in bold.

F p

S1 F p

0.77 0.35 0.34

0.67 0.52

0.77

S2 F p

0.79 0.26 0.15

0.49 0.22

0.82

S3 F p

0.30 -0.04 0.28

0.23 0.08

0.50

S4 F p

0.58 0.15 0.40

0.22 0.08

0.63

S5 F p

0.53 0.22 0.34

0.16 0.14

0.64

In all cases, the correlations were carried out over

subsequent, non overlapping windows of 2000 points

each, from signals of about 64,000 points sampled

at 256 Hz. The aim was to check whether in the

time spanned by the signal, namely within the about

138 sec of its total duration, some signiﬁcant spec-

tral change occurred. An even more ambitious goal

was to enlight a space-dependent trend linked to the

rostro–caudal direction.

Although the data in Table I did not substantiate

clearly the above expectations, it seems fair drawing,

on their basis, the following minimal conclusions: a)

all the analyzed signals show a quite synchronous be-

haviour, between the left and right emispheres, in the

fronto-parietal and occipital lobes; b) the signal as-

sociated to the migraine diagnosis shows the highest

correlation as compared to both the epileptic cases;

c) the highest synchronous activity is concentrated

in the occipital lobe under all conditions. Moreover,

the concomitant lower and higher synchronization in

the temporal/central and frontoparietal areas, respec-

Figure 2: Electrode location on the human scalp in the ”10-

20 montage”. According to a standardized procedure 10

small electrodes (less than 1 cm diameter, wetted by a salt-

past increasing the electrical conductivity) are symmetri-

cally located on each emisphere according to the following

nomenclature: Fp = frontal-pole; T = temporal, C = central,

P = parietal, O = occipital.

Figure 3: Oscillating activity of brain neurons simulated by

a MAS (Wilensky, 2009). Left and right panels refer to

the main peaks of activity within a single oscillation cycle.

Each cycle lasted about 10 machine time units, correspond-

ing to about 1 s, and involved 1500 agents in both hemi-

spheres, whose behaviour was initially random. The oscil-

lating regime apparent in the two panels arises after about

350 machine time units. The algorithm used in the sim-

ulation has been described in detail elsewhere (Colosimo,

2008)

tively, are consistent with an oscillating behaviour,

namely a clustering in well deﬁned areas of the maxi-

mal and minimal activity occurring in the considered

time span. It is worth mentioning that a similar (al-

though less clear) trend is also observed by correlat-

ing EEG records from proximal electrodes within the

same (left) hemisphere, reported in columns 4,5 of

Table I. The data concerning the other (right) hemi-

sphere are almost the same.

EEG SIGNALS IN EPILEPSY AND MIGRAINE - Analysis and Simulations by Multi-agent Systems

449

Figure 4: A MAS based oscillating ﬁeld model for simulating hypersynchronized neuron waves. In this case the two brain

hemispheres have been divided into four identical quadrants and the fraction of the total (5,000) neurons active in one the

quadrants reported.

3.2 Simulating Cortical Spreading

Depression

The ﬁrst conjecture about the causal relationships

linking synchronization and epilepsy dates back to

Matsumoto (1964),(Matsumoto and Ajmone-Marsan,

1964) showing that hyperactivity of a limited num-

ber of cells unable to recruit a larger network was

also unable to originate an epileptic event. The some-

how paradoxical discovery of the extensive synchro-

nization occurring in migraine is due to Leao (Leao,

1944), while studying an epileptic model in rabbits.

Leao observed a depolarizing wave moving at a 3

mm/min speed in the rabbits cortex. He named the

wave Cortical Spreading Depression (CSD), since af-

ter its passage the cortex remained inactive for some

time. Only in 1994, however, Lauritzen (Lauritzen,

1994) hypothesized that CSD could have been at

the origin of the visual aura in human migraine.

He showed that associated with the visual aura was

a high-activity wave moving in the anterior direc-

tion from the occipital region at speed from 2 to 6

mm/min. Such a wave was followed by a tempo-

rary suppression of the cortical electrical activity. The

frequent absence of the visual aura in many subjects

has been explained by assuming that CSD may also

originate in visually silent regions (Pietrobon, 2005).

CSD, in fact is not limited to the occipital area: its

starting point may be observed most frequently in the

CA1 hyppocampal area, followed by the neo-cortex,

and it remains a most interesting phenomenon of neu-

ral synchronization.

Figure 3 shows the activity patterns observed in

the area representing a coronal section of the human

brain, by means of a simulation device described else-

where (Wilensky, 2009). The 2 panels in the ﬁgure

show the clustering of active neurons in different re-

gions of the ”brain” during a repeating functional cy-

cle. The period of such cycle can be easily modulated

by a number of factors, primarily of metabolic nature.

4 CONCLUSIONS

Even if neither migraine nor epilepsy are actually

fully understood in their deep causes and detailed

mechanisms, a most probable connection between

them concerns the abilitiy of neural cells to get patho-

logically hypersynchronized under various circum-

stances. In this frame, it is maybe worth stressing

that the main goal of our research plan, is to simu-

late the cortical depression wave dynamics. Although

obviously related to the synchronized activity of neu-

ron populations, this represents a higher level of com-

plexity, since it involves both a time and space depen-

dence of the oscillatory activity whose reproduction

in silico, at our knowledge, has not been successfully

attempetd as yet.

All in all, the most interesting outcome of our

study may be summarized as follows:

- Concerning the analysis of EEG signals, a nec-

essary prerequisite to any modelistic effort, a simple

and ﬂexible tool like the Pearson correlation coefﬁ-

cient showed considerable heuristic power: as a mat-

ter of fact Figures 3,4 and Table I indicate that by

just dissecting the time series into a number of subse-

quent windows in order to increase the resolution of

the method, allowed to identify the presence of time

and space ordered activity patterns of neurons from

both homo- and contralateral signals.

- Concerning the multi-agent simulation environ-

ment, NetLogo (Wilensky, 2009)appeared more ﬂexi-

ble as compared to other programmable tools special-

ized for neuronal systems, like, for example, Gene-

BIOSIGNALS 2010 - International Conference on Bio-inspired Systems and Signal Processing

450

sis (Beeman and Bower, 2009) or Neuron (Carnevale

and Hines, 2006), although probably less powerful

in terms of manageable models sizes. (Brette et al.,

2007) As an example, by the very same tool (Netlogo)

it was relatively straightforward to work out simula-

tions as those shown in ﬁgures 3 and 4.

- Although still far from conclusive, our results

and, in particular, the similarity of the simulated sig-

nals in ﬁgure 4 with the alternating bursts of activities

and ’interictal’ phases, observed in vitro (Panuccio

et al., 2009) and in vivo (Steriade, 2006), represents

an encouraging ﬁrst step towards the clariﬁcation of

neural pathologies by means of relatively simple and

ﬂexible numerical methods.

ACKNOWLEDGEMENTS

We would like to thank all the colleagues from the

CISB InterDept. Center of Sapienza University for

suggestions, discussions and encouragement. AV is

grateful to Lorenzo Rocchi for his help on everyday

EEG practice.

REFERENCES

Beeman, D. and Bower, J. (2009). The GENESIS Simula-

tion System. Web Site. http://genesis-sim.org.

Brette, R., Rudolph, M., Carnevale, T., Hines, M., Beeman,

D., Bower, J., et al. (2007). Simulation of networks

of spiking neurons: a review of tools and strategies. J

Comput Neurosci, 3(23):349–98.

Carnevale, N. and Hines, M. (2006). The NEURON Book.

Cambridge UK: Cambridge University Press.

Colosimo, A. (2008). Biological simulations by au-

tonomous agents: two examples using the NetLogo

environment. Biophysics and Bioengineering Letters,

1(3):40–50.

Flink, R., Pedersen, B., Guekht, A., Malgren, K.,

Michelucci, R., Neville, B., and et al (2002). Guide-

lines for the use of EEG methodology in the diagno-

sis of epilepsy. International League Against Epilepsy:

commission report. Commission on European Affairs:

Subcommission on European Guidelines. Acta Neurol

Scand, (106):1–7.

Hunter, M., Smith, R., Hyslop, W., Rosso, A., Gerlach, R.,

Rostas, J., Williams, D., and Henskens, F. (2005). The

Australian EEG Database. Clinical EEG and neuro-

science, 36(2).

Lauritzen, M. (1994). Pathophysiology of the migraine

aura. The spreading depression theory. Brain,

1(117):199–210.

Leao, A. (1944.). Spreading depression of activity in the

cerebral cortex. J. Neurophysiol, 1(7):359–90.

Lipton, R. and Bigal, M. (2005). The epidemiolgy of mi-

graine. The American Journal of Medicine, 118(Supp

1):3s–10s.

Mathworks (2006). MatLab – The language of technical

computing. Web Site. http://www.mathworks.it/.

Matsumoto, R. and Ajmone-Marsan, C. (1964). Cortical

cellular phenomena in experimental epilepsy: ictal

manifestations. Exp Neurol 1964, (80):305326.

Ottman, R. and Lipton, R. (1996). Is the comorbidity of

epilepsy and migraine due to a shared genetic suscep-

tibility? Neurology, 4(47):918–24.

Panuccio, G., Curia, G., Colosimo, A., Cruccu, G., and

Avoli, M. (2009). Epileptiform synchronization in the

cingulate cortex. Epilepsia, 3(50):521–36.

Pietrobon, D. (2005). Migraine: new molecular mecha-

nisms. Neuroscientist, 4(11):373–86.

Rodgers, J. and Nicewander, W. (1988). Thirteen ways

to look at the correlation coefﬁcient. The American

Statistician, 42(1):59–66.

Rogawski, M. (2008). Common pathophysiologic mech-

anisms in migraine and epilepsy. Arch Neuro,

6(65):709–14.

Russell, S. and Norvig, P. (2005). Artiﬁcial Intelligence: a

modern Approach. Prentice Hall International.

Sack, O. (1992). Migraine. Random House Inc.

SAS (2007). SAS - Home page. Web Site.

http://www.jmp.com.

Steriade, M. (2006). Grouping of brain rhythms in corti-

cothalamic systems. Neuroscience, (137):1087–1106.

Syvertsen, M., Helde, G., Stovner, L., and Brodtkorb, E.

(2007). Headaches add to the burden of epilepsy. J

Headache Pain, (8):224–230.

Wilensky, U. (2009). NetLogo - Home page. Web Site.

http://ccl.northwestern.edu/netlogo/.

EEG SIGNALS IN EPILEPSY AND MIGRAINE - Analysis and Simulations by Multi-agent Systems

451