Statistical Analysis of the Human EEG during RF

Exposure from Mobile Phones: An Alternative Method

to Analysis of the EEG in Frequency Bands

Howard D’Costa

1, 2

, Irena Cosic

1, 2

School of Electrical & Computer Engineering, RMIT University, Melbourne, Australia

Australian Centre for Radiofrequency Bioeffects Research

Abstract. This paper aims to describe a novel statistical approach to analysing

the effects of radiofrequency (RF) exposures from mobile phones on the human

EEG. In addition, the paper describes two limitations that may be encountered

when using statistical methods to analyse the EEG in its frequency bands. The

proposed method of analysis which is based on measures of central tendency

introduces an approach whereby the recorded body of EEG data collected

during trials can be effectively interpreted for spectral analysis at a higher

resolution across the EEG spectrum. It is believed that the proposed statistical

approach may be also useful in other studies investigating the effects of

alternate forms of involuntary stimulus on the human EEG, such as electrical

stimulus, light, and sound.

1 Introduction

It is as yet undetermined whether mobile phone exposures can cause adverse health

implications or changes in human brain function. In an attempt to address these

concerns, researchers have utilised electroencephalographic (EEG) recordings to

determine whether radiofrequency (RF) emissions from mobile phones influence

human brain wave activity. A common approach to statistical analysis in these

investigations, is to analyse the EEG in its generally classified frequency bands,

namely delta (1-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), and beta (13-32 Hz)

[1][2][3][4]. Nevertheless, amongst these studies there can be found slight variations

in the EEG spectral ranges analysed. In a study by Hietanen et al.[5] EEG recordings

were obtained from 19 participants during exposure in separate tests to five active

mobile phones operating at either 900 MHz or 1800 MHz. The phones were

positioned 1 cm from the left side of the head and generated a peak output power

ranging from 1 - 2 W. With the exception of the delta region from one of the test

phones statistical comparisons drawn between control and exposure trials indicated no

significant changes in the EEG frequency bands 1.5 – 3.5 Hz (delta), 3.5 – 7.5 Hz

(theta), 7.5 – 12.5 Hz (alpha), and 12.5 – 25 Hz (beta). In another investigation by

Reiser et al. [6] 36 subjects were exposed to a mobile phone’s RF emissions for a

duration of 15 minutes. The mobile phone had a carrier frequency of 902.4 MHz,

which was modulated at 217 Hz. The phone was programmed to transmit at 8 W, and

D’Costa H. and Cosic I. (2005).

Statistical Analysis of the Human EEG during RF Exposure from Mobile Phones: An Alternative Method to Analysis of the EEG in Frequency Bands.

In Proceedings of the 1st International Workshop on Biosignal Processing and Classiﬁcation, pages 167-174

DOI: 10.5220/0001197301670174

 SciTePress

was placed at a distance of 40 cm from the rear of the head during the experiment.

Results of the study indicated power increases in the EEG frequency bands of 9.75 Hz

– 12.5 Hz (alpha 2), 12.75 – 18.5 Hz (beta 1) and 18.75 – 35 Hz (beta 2). The

increases occurred approximately 15 minutes after exposure ceased.

Although investigations in this area of study have until now concentrated on analysis

of the EEG in its spectral bands, there is however significant limitations to this

approach that should be considered. With respect to the utilisation of the EEG as

basis to detect an external stimulus, these limitations primarily arise from the

moderate spectral resolution analysis imposed by relatively wide ranges of the EEG

frequency bands. As opposed to analysis of the EEG in spectral bands, alternate use

of non-linear statistical methods have been produced by others in related

electromagnetic field effect studies [6][7].

From an adapted analysis of our previous work (D’Costa et al. [8]), this paper aims to

present a novel statistical approach to analysing the human EEG where all frequencies

within the EEG spectrum can be analysed. In addition, the paper aims to describe and

outline the limitations associated with statistical analysis of the EEG in its spectral

bands.

2 Limitations of Analysis of the EEG in Frequency Bands

There are two evident limitations associated with statistically analysing the EEG in its

frequency bands for the purpose of determining whether an external stimulus such as

mobile phones affect human brain waves. These limitations may be described as

follows:

1. Important data is potentially lost due to averaging in frequency bands when

drawing comparisons between control and exposure EEG data sets. For example, the

alpha EEG band spans over five distinct frequencies from 8 -13 Hz. In order to

prepare this band for hypothesis testing the total EEG power across each of the five

frequencies must be averaged to one value for both the exposure and control test

recordings. For this reason it is arguable that an effect due exposure can occur in any

one of the five frequencies though may be lost through averaging. The probability for

this loss occurring is even more so for the beta band (>13 Hz) where up to 20 or more

frequencies may be averaged.

2. Identification of potential changes in frequency ranges spanning across the EEG

band divisions are not observable which may mask potential effects. For example if

an alteration in the EEG due to an applied exposure existed over a range spanning

from 5 – 9 Hz the effect may become impossible to observe as theta (4 – 8 Hz) and

alpha (8 - 13 Hz) must be independently analysed.

An example of results for an analysis conducted in frequency bands is shown below

in Table 1 (adopted from D’Costa et al. [8]). In this study ten participnats were

exposed to a mobile phone operating at 900 MHz at nominal full-power (2 W peak

output). The EEG was recorded from the frontal, central, and occipital regions of the

head in a series of five control and five exposure tests. A paired t-test analysis was

conducted to draw statistical comparisons between the averaged control and exposure

168

test recordings. The t-test results are indicated below for the four EEG frequency

bands of interest analysed.

Table 1. Shows an example of results adopted from D’Costa et al.[8] for an analysis conducted

in the EEG frequency bands delta (1-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), and beta (13-32

Hz). Statistical levels were considered significant at p-values < 0.05 (shown in italic)

It can be observed in table 1 above that the t-tests results indicated statistically

significant differences in the alpha and beta bands (p-values < 0.05). However, in

contrast and based on the limitations described above, it may also be shown that it is

not possible to determine whether the mobile phone exposure produced a potential

influence across or at particular frequency rhythms within the four EEG bands.

3 Proposed Analysis of the EEG

By employing existing statistical methods, this section describes a novel approach

where all frequencies in the EEG spectrum are considered to investigate the effects of

mobile phone RF exposures on human brain wave activity. The method of analysis is

described in the following 3 stages given the basic case that four control EEG

recordings are to be compared to four exposure EEG recordings acquired from a

sample size of 10 participants (EEG spectral range is 1-32 Hz):

Step 1. From each of the four control and four exposure recordings 4 × 32 EEG

power values (×10) are generated. For each participant, the median EEG power value

of the four control recordings in each EEG frequency (from 1 -32 Hz) minus the

corresponding median in the four exposure recordings is calculated. The resultant

number of positive values or decreases in each EEG frequency is then identified over

the entire sample. Table 2 below demonstrates a mock example of an output table

generated for this step at the arbitrary rhythm of 7 Hz. In a similar manner, the

number of negative values or increases may be alternatively chosen.

Full-power mode trial

Delt

het

ha Bet

Recording

Site

95% CI

(µV)

p-value 95% CI

(µV)

p-value 95% CI

(µV)

p-value 95% CI

(µV)

p-value

Frontal

-2.3, 7.0 0.281 -1.3, 3.8 0.289 -0.5, 1.8 0.264 -0.6, 1.1 0.519

Central

-1.0, 9.0 0.106 -1.9, 7.0 0.232 0.1, 3.7 0.038 0.03, 1.9 0.045

Occipital

-0.4, 11.3 0.065 -1.3, 8.8 0.13 -0.1, 5.3 0.06 0.01, 3.0 0.049

169

Table 2. Shown is an example of the paired sample of median values of the control and

stimulus EEG recordings at 7 Hz. The median differences in EEG power and respective

identified decreases in EEG can be observed. In this example 8 out of the 10 participants

indicated a decrease in the median EEG power at 7 Hz

By observation of Table 2, it can be seen that up to 80% of the participants for

example indicated a decrease in EEG power at 7 Hz. By repeating this process for all

32 frequencies a distribution of the percentage of subjects indicating a decrease in

EEG power versus frequency may be generated. If EEG recordings were to be

simultaneously acquired from multiple recording site locations on the head a 3D

illustration of these distributions may be shown as given for example in Figure 1(a).

Participant

Median value of

control recordings

at 7 Hz (µV)

Median value of

stimulus recordings

at 7 Hz (µV)

Median

difference

at 7 Hz (µV)

-1

-2

Noted

decrease

in EEG

Fig. 1.(a) Shows an example of distributions of the percentage of subjects indicating a decrease

in EEG power versus frequency as generated in step 1. As shown, distributions produced for

simultaneous EEG recordings acquired over multiple recording site locations may be indicated.

In this example three recording sites are considered over the frontal, central, and occipital

regions of the head (example adopted from D’Costa et al.[8])

Recording

Site

EEG Rhythm

(Hz)

% of Subjects

Indicating

Decrease in

EEG Power

Frontal

Central

Occipital

100

(a)

170

Fig 1(b) The corresponding distribution plots of figure 1(a) as generated for step 2. The

‘boxes’ indicate where at least 50% of the distribution lies. The ‘box-whiskers’ and asterisks

(data outliers) together indicate where at least 25% of the distribution lie

Step 2. For further interpretation, the distributions generated in step 1 (Fig. 1(a)) are

interpreted to standard Box Diagram plots as indicated in Figure 1(b). As per

standard [10] the ‘boxes’ indicate where at least 50% of the distribution lies. The

‘box-whiskers’ and asterisks (data outliers) together indicate where at least 25% of

the distribution lie.

Distribution plots for a larger number of recording regions may also be effectively

shown as Figure 2 below indicates for a standard 16 -point EEG recorded array (10-

20 International standard [11]).

OccipitalCentralFrontal

% of Subjects

Indicating

Decrease in

EEG Power

Recording Site

(b)

171

Step 3. Lastly, with reference to step 1, a paired t-test analysis is conducted on the

median paired data sets in each frequency where 75-100% of subjects indicate an

increase or decrease in EEG power. Table 3 below demonstrates an example of an

output table of results that may be generated for this step. The table can be observed

to indicate 95% confidence intervals and corresponding p-values produced by the

paired t-test analysis for each of the identified frequency rhythms. The p-values in

this example are considered to be statistically significant for values less than 0.05 as

indicated in italic (example adopted from D’Costa et al.[8]).

Larger data sets, such as in the case for a 16-point EEG analysis, may also be

represented in this manner, or simplified to indicate the significant statistical values.

Table 3. Shows an example of a resultant output table generated in step 3 above

1. Significant p-values

indicated at 9 Hz from

all 3 recording sites

2. Significant p-values

indicated at 7 & 9 Hz

from both the central

and occipital regions

4. Significant p-values

indicated at consecutive

frequencies at 17 & 19

Hz from the occipital

region

3. Significant p-values

indicated at consecutive

frequencies at 7, 8, & 9

Hz from the occipital

region

EEG

Rhythm

(Hz)

Rec.

Site

% of

Subjects

Indicating

Decrease

95% CI

(µV)

p-value

3 F 20 -3.1, 4.6 0.673

9 F 90 0.1, 0.6 0.009

3 C 80 -1.9,10.2 0.155

5 C 80 -1.3, 7.5 0.141

7 C 80 0.6, 5.0 0.020

9 C 80 0.5, 4.9 0.022

19 C 80 -0.1, 1.7 0.076

30 C 80 -0.1, 1.7 0.070

4 O 80 -0.9,10.3 0.089

5 O 80 -0.8, 9.7 0.087

7 O 90 0.5, 9.0 0.032

8 O 90 0.7, 8.7 0.026

9 O 80 0.9, 6.9 0.017

10 O 80 -0.4, 7.5 0.073

14 O 80 -0.3, 5.2 0.071

16 O 80 -0.2, 5.7 0.065

17 O 80 0.0, 5.3 0.050

19 O 80 0.0, 4.9 0.050

20 O 80 -0.4, 5.4 0.086

22 O 80 -0.1, 4.3 0.063

23 O 80 -0.1, 3.6 0.056

25 O 90 -0.1, 3.5 0.061

26 O 80 -0.2, 3.0 0.082

27 O 90 0.1, 2.6 0.034

28 O 90 -0.0, 2.4 0.055

29 O 90 -0.2, 2.4 0.078

30 O 80 -0.5, 2.7 0.151

31 O 80 -0.7, 2.4 0.248

32 O 80 -0.5, 1.8 0.231

Mobile phone trial

172

4 Discussion

The proposed method of analysis can be useful in determining whether there is an

effect in the EEG due to mobile phone exposures for several reasons. In general, the

analysis uses an approach by which the raw EEG data is used to identify where

probable effects may occur. To do this, firstly the percentage of subjects indicating a

decrease or increase in each EEG frequency is identified as described in step 1. It

follows in this step that a resultant distribution plot is generated as a function of

percentage of subjects indicating a decrease in EEG power versus frequency (Fig.

1(a)). To interpret this figure clearly it is hypothesised that if there is no change in the

EEG due to the mobile phone exposure over a given sample size, the distribution at

each recording site should tend towards being a uniform 50% over the EEG spectrum.

It thus follows the more the percentage of subjects indicating a change in EEG power

tends away from 50% the more probable it is that a significant difference occurs in

those EEG frequencies demonstrating higher and lower tendencies. Consequently, it

is of interest in step 3 to test the statistical significance of difference in these rhythms.

In addition to the generated distributions in step 1, step 2 introduces the use of box

diagram plots (Fig. 1(b), Fig. 2). As may be observed, the diagrams represent where

the corresponding distributions produced in step 1 lie with respect to each other and

their respective recording sites. In particular, this characteristic is very useful in

demonstrating how the frequency distributions of recording sites near to the position

of a mobile phone RF source may vary with distance.

In the final third step, it is described that a paired t-test analysis is conducted on the

median paired data sets in each frequency (Table 2) where 75-100% of subjects

indicate an increase or decrease in EEG power. By examining this upper high

tendency range (away from 50%) this important stage of analysis significantly

reduces analysing large proportions of probable redundant data and concentrates on

interpreting more likely affected regions. Statistically significant results determined

in this step would therefore be more difficult to disregard as occurring due to

statistical chance. Demonstrated in Table 3 is an example of a resultant output table

generated from step 3. The table shows four prominent trends which occurred

amongst frequencies indicating statistically significant differences in the median

control and exposure EEG recorded sets. Results of prominent interest indicated in

this example from our previous work [8] were EEG frequencies showing statistically

significant changes in EEG power from the occipital region at 7 Hz, 8 Hz, and 9 Hz.

It is of particular interest to note for the purpose of this work that this potentially

important result indicating significant change in consecutive rhythms extending from

within the theta to the alpha EEG range may have otherwise been masked by an

analysis in frequency bands due to the two limitations earlier discussed.

5 Conclusion

This paper proposes an alternative method to analysing the effects of mobile phone

exposures on the EEG in its distinct frequency bands. The main advantage of the

proposed analysis is that all frequencies within the EEG spectrum are considered

173

resulting in a higher resolution analysis to detect potential stimuli from exposure. It is

important to point out in such a case where the effects of an external stimulus is of

interest that it is not important to conduct analysis of the human EEG in its frequency

bands. This is due to the fact that the power in the EEG frequency bands is a

physiologically and mentally dependent parameter that presumably does not differ

during control and exposure conditions. Thus for linear analysis the EEG may be

statistically handled in a manner whereby it is fixed.

Overall, it is thought that the proposed spectral analysis of the EEG is a robust and

sensitive method for which to investigate the effects of radiofrequency exposures

from mobile phones on the human EEG. We look forward to incorporating and

further adapting this method into our current study that is underway to examine the

effects of GSM mobile phone exposures on multiple biosignal responses.

Acknowledgements

We would like to extend our thanks to Dr. Everarda Cunningham for her kind support

in reviewing this work.

References

1. Cook, C.M., Thomas, AW., Keenliside, L., Prato, F.S.: Resting EEG Effects During

Exposure to a Pulsed ELF Magnetic Field. Bioelectromagnetics (2005) 26(5):367-76

2. Eulitz, C., Ullsperger, P., Freude, G., Elbert, T.: Mobile phones modulate response patterns

of human brain wave activity. Neuroreport (1998) 9: 3229-3232

3. Huber, R., Graf, T., Cote, K., Wittmann, L., Gallmann, E., Mather, D., Schuderer, J.,

Kuster, N., Borbely, A., Achermann, P.: Exposure to pulsed high-frequency

electrogmagnetic field during waking affects human sleep EEG. Neuroreport (2000) 11:

1321-1325

4. Röschke, J., Mann, K.: No short-term effects of digital mobile radio telephone on the

awake human electroencephalogram. Bioelectromagnetics (1997) 18: 172-176

5. Hietanen, M., Kovala, T., Hämäläinen, A-M.: Human brain activity during exposure to

radiofrequency fieldsemitted by cellular phones. Scandinavian Journal of Work

Environment & Health (2000) 26: 87-92

6. Bachmann, M., Kalda, J., Lass, J., Tuulik, V., Säkki, M., Hinrikus, H.: Non-linear analysis

of the electroencephalogram for detecting effects of low-level electromagnetic fields.

Medical & Biological Engineering & Computing (2005) 43:142-149

7. Lipping, T., Olejarczyk, E., Parts, M.: Analysis of photo-stimulation and microwave

stimulation effects on EEG signal using Higuchi’s fractal dimension method. Proceedings

of SPIE (2004) 5505: 174-178

8. D’Costa, H., Trueman, G., Tang, L., Abdel-rahman, U., Abdel-rahman, W., Ong, K., Cosic,

I.: Human brain wave activity during exposure to radiofrequency field emissions from

mobile phones. Australasian Physical & Engineering Sciences in Medicine (2003) 26[4]:

162-167

9. Reiser, H.-P., Dimpfel, W. Schober, F.: The influence of electromagnetic fields on human

brain activity. European Journal of Medical Research (1995/96) 1: 27-32

10. Sirkin, R. Mark.: Statistics for the Social Sciences. 2nd Edn. Thousand Oaks, Calif.:Sage

Publications (1999)

11. Aston R, Principles of Biomedical Instrumentation and measurement, Meril (1990)

174