Вы находитесь на странице: 1из 6

A new EEG recording system for passive dry electrodes

Objective: We present a new, low power EEG recording system with an ultra-high input impedance that enables the use of long-
lasting, passive dry electrodes. It incorporates Bluetooth wireless connectivity and is designed to be suitable for long-term
monitoring during daily activities. Methods: The new EEG system is compared to a standard and clinically available reference
EEG system using wet electrodes in three separate sets of experiments. In the first two experiments, each dry electrode was
surrounded by four standard wet electrodes and the alpha and mu-rhythms were recorded. In the third experiment, serial
monopolar (referred to the left ear) recordings of flash visual evoked potential were performed using the new EEG system and a
reference system. Results: These experiments showed that the signal recorded using the new EEG system is almost identical to
that recorded with standard clinical EEG equipment; our measurements showed that the correlation coefficient between the dry
electrode recordings and the average of the four standard electrodes surrounding each dry electrode is greater than 0.85.
Conclusion: We conclude that the new EEG system performs similarly to reference EEG systems, while providing the advantages
of portability, ease of application and minimal scalp preparation. Significance: The proposed system using passive dry
electrodes suitable for single use while performing as good as standard EEG equipment provides ease of application and
minimal scalp preparation. _ 2009 International Federation of Clinical Neurophysiology. Published by Elsevier Ireland Ltd. All
rights reserved. .

1. Introduction

In this paper we present a new system (hardware using passive dry electrodes) for EEG recording and compare
it to a standard commercial EEG recording system using conventional wet electrodes. The EEG recording
system consists of (1) an ultra-high input impedance bio-amplifier; (2) new dry electrodes suitable for long-term
recording in real environments; and (3) wireless connectivity using a very low power ADC equipped Bluetooth
module (approved for medical devices). The EEG recording system forms part of a larger and joint project
between the University of Sydney and Federico II University of Naples to develop hardware and software for
an easy to use, wearable BCI (braincomputer interface) system that is referred to as Penso. In this paper we
focus on the evaluation of the hardware system.

The standard procedure for application of wet electrodes and performing EEG recordings requires substantial
preparation and the application of conductive gels and glues (Teplan, 2002). The preparation typically consists
of head measurement, accurate electrode placement, and skin preparation which may involve an abrasive
paste or soap (Teplan, 2002). Single electrodes are often kept in position using collodion. This method of
electrode placement allows reliable, stable, and repeatable EEG recordings. It is recommended practice to
maintain the electrode contact impedance below 5 k X and any impedance imbalance to within 1 kX. Longterm
recordings pose their own unique set of challenges that include the desiccation of the conductive gel or paste
that in turn decreases signal-to-noise ratio and increases contact impedance together with sweat and gel
leakage that can cause electrical short circuits between adjacent sites.

The use of dry electrodes may be a possible solution for stable long-term EEG recordings and increased ease of
use. There are a number of advantages that our dry electrode system offers. Firstly, it does not require
conductive gel or paste between the skin and the electrode surface to record a signal. Furthermore, it does not
require any skin/scalp preparation. These results are achieved by using a bio-amplifier with an unusually high
input impedance and careful shielding. We have developed a simple methodology to apply our flexible, dry and
flat electrode to the scalp without the need to shave hair. It is thought that an advantage of the wet electrodes
system is the application of a conductive gel that allows contact of the electrode with the scalp without the
need to remove scalp hair. However, the same feat is accomplished with our dry electrode system without the
use of a conductive gel. Of course, shaving may provide some improvements in performance, but none of the
subjects for in our experiments were shaved.

2. Methods

2.1. The hardware

Fig. 1 shows a block diagram of the EEG recording system. The monopolar signals measured by the electrodes
are referred to a common electrode placed on the left ear lobe (the reference electrode). A grounding electrode
is placed on the right ear lobe. The equipment includes special circuitry for the prevention of microshocks
(Webster, 1998). All of the circuitry is battery powered and floating with respect to ground, while the leakage
current on each electrode is limited to 200 nA to ensure that the device is safe for use. Careful shielding is
adopted to reduce EM interference on the electrode cables and the pre-amplifiers. The signals are digitized at
12 or 16 bits after amplification and filtering and transmitted to a PC via Bluetooth. The measured input
referred noise for each channel is less than 2 lVpp in the bandwidth up to 10 Hz.

The dry electrodes are made with commercially available 1.5 mm thick silicone conductive rubber in the form of
discs of 8 mm diameter. This material has been used in electrodes for decades, but mainly to make stimulation
electrodes because of its intrinsic high ohmic resistance (Searle and Kirkup, 2000; Artz, 1970). Fig. 2 shows an
illustrative diagram of the dry electrode. In practice a final layer of insulation is added to cover the shielding
plate to protect it from contact. The active side of the electrode is capacitively coupled through a layer of
insulating silicon rubber with a metal shield connected to the active guard shield. Power is supplied by a high
capacity (4.8 V) NiMh battery (similar to a mobile phone battery) regulated at two different voltage supply
values. A 3.3 V supply is used to power the digital Bluetooth circuitry and a 3.6 V supply is used to power the
analog side of the Bluetooth transceiver chip which contains a software configurable analog multiplexer and an
analog to digital converter (ADC). The full battery voltage (4.8 V) is used to power the EEG front-end circuitry.
The maximum power consumption of the circuit with eight analog channels sampling at 128 Hz and the
Bluetooth transceiver operating at maximum power is below 250 mW, providing several hours of continuous
operation at maximum power consumption. Operating the Bluetooth receiver in SNIFF mode and sampling data
at 500 Hz, the power consumption drops below 50 mW. For prac- tical long-term monitoring, multiple batteries
would have to be used with recharging (Gargiulo et al., 2008).

2.2. The comparison experiments

Evaluation of EEG systems can be difficult, since many parameters are involved, and EEGs cannot be
reproduced in different recording sessions. The evaluation of new EEG recording systems is normally conducted
by comparison with a reference system using two methods. In one method, referred to as the parallel method,
electrodes for the test and reference EEG system are placed at neighboring locations on the scalp and the
signals are examined simultaneously in the time and frequency domains (Iguchi et al., 1994). In another
method, referred to as the serial method, the comparison between the reference and test system is made
serially by first connecting the reference system to a subject who performs a particular task and then repeating
this measurement with the test system (Popescu et al., 2007). Our approach combined both techniques in three
experiments. We used the parallel method to recorded simultaneously EEG signals for a (i) reactive alpha
rhythm, and (ii) the mu-rhythm while performing a braincomputer interface (BCI) task. In addition, we used the
serial method in a third experiment to record EEG signals from the same position on the scalp, in different trials
during a reproducible task associated with flash visual evoked potentials (FVEP).

Prior to each comparison experiment, both systems were examined using a signal generator to determine
hardware features such as filter cut-off frequencies and gain. An exact match of these hardware features was
neither expected nor possible due to the presence of various hardware notch filters and varying filter responses
(see Table 1). In addition, we note that an experienced electroencephalographer examined the recorded EEG
signals for the three presented experiments; thus we can say that the EEG signals recorded from all the
subjects, from both the test and reference EEG systems demonstrated normal brain activity in awake subjects.

We first used the parallel method to record EEG signals. The test EEG system was configured with dry
electrodes placed in the standard positions C3, C4 and Cz. The reference system (Compumedics _, Abbotsford,
Victoria, used at Royal Prince Alfred Hospital, Sydney, NSW) recorded simultaneously from standard locations
Fc3, Fc4, Fcz, C1, C2, C5, C6, PC3, PC4 and PCz as depicted in Fig. 3. The reference EEG system used standard
golden brass cup electrodes with conductive gel (wet electrode) held in position using with collodion. Each dry
electrode was surrounded by four wet electrodes.

These experiments were performed on eight untrained healthy volunteers (seven male one female, age range
between 25 and 45 years old). Approval was obtained from the Ethics Committee at the University of Sydney
and informed consent was obtained from each subject. It is worth to highlight that electrodes were placed by a
researcher who was trained for a week at the RPA hospital on how to identify electrode positions using head
measurement, and on skin preparation for golden brass electrodes.
In the first part of the experiment we evaluated the alpha rhythm (Mohanan and Rathinam, 1995). This is
because the alpha rhythm is easily recognizable and is one of the strongest brain signals in the EEG (Webster,

The mu-rhythm was also studied because of its unique characteristics. In the second part of the experiment the
subjects were asked to perform a BCI 1D cursor control task (leftright movement) and generating a mu-
rhythm. Random left and right targets were presented to the subject from the BCI computer (connected to our
system) and the cursor movements were governed by the murhythm power spectra of the electrodes C3, C4
and Cz. The experiment was divided into three separate groups of trials. Trial 1 was a familiarization trial of
approximately 3 min in which the subjects were asked to press a button with either their left or right hand
depending on the position of the target. In Trial 2, a pre-BCI trial which lasted approximately 36 min, the
subjects were asked to imagine the pressing of a button with either their left or right hand depending on the
position of the target. In Trial 3, the subjects were asked to try to control the position of a cursor using a BCI
system that monitored the power spectra of the C3, C4 and Cz electrodes. This trial generally continued until
the subject became tired, which was generally after about 15 min. EEG signals were recorded during all of the
trials and the data collected from all of the trials were used for analyses.

The EEG signals recorded with both systems were band-pass filtered (bandwidth 0.535 Hz; 50th order FIR) and
an additional 50th order IIR notch filter at 50 Hz was applied for the first two comparison experiments; see Table
1 for further details. Because the test and reference EEG systems were controlled using separate computers, a
precise time alignment between the test and reference EEG signals was required. Time alignment was obtained
based on the maximum of the correlation between signals recorded from each dry electrode and the average of
the four surrounding electrodes.

The third comparison experiment involved recording a visualevoked response. The visual evoked potential is a
transient response elicited by specific visual stimuli. It is used in combination with continuous EEG recordings to
document the integrity of neural pathways and does not require a motor response (Teplan, 2002). An example
of such elicited responses is the flash visual evoked potential (FVEP). Despite the fact that FVEPs are very
variable (Teplan, 2002), they are often used in braincomputer interface applications (Lee et al., 2005; Wu et
al., 2004). In this experiment we compare the test EEG system configured with dry electrodes with a reference
system referred to as the g.BSamp system which used standard brass golden cup electrodes. The g.BSamp
system is in use at the biomedical lab at the University of Naples. As previously, both systems were first tested
using a signal generator (see Table 1. Specification of the different equipment used) Also as previously, the EEG
signals recorded with both systems were filtered using a software band-pass filter, 0.05100 Hz (50th order FIR)
and an additional 50th order IIR notch filter at 50 Hz was applied.

The experiment was conducted following standard FVEP recommendations (Odom et al., 2004). We recorded
FVEP from three subjects. All subjects were male, healthy, and without any known visual or mental
impairments. They had normal visual acuity and did not require refractive glasses and their age ranged from 24
to 30 years. For this experiment electrodes were placed at standard position Oz. An additional ground electrode
was placed on the right ear lobe. The recording positions were marked on the head using an eye-liner pencil
and our dry electrodes were secured in position using an elastic band to avoid entangling the hair. The subjects
were asked to sit in an armchair for 30 min in a dimly lit room to allow light adaptation. After this period a flash
of light in front of the subject would produce the necessary stimuli as previously described (Odom et al., 2004).

The stimuli were randomly presented by the operator with an average interval of 2 s. Signals from the flash
trigger and a light detector were recorded simultaneously with the EEG signal. The starting time for each trial
was defined as 1 ms after the flash trigger. Consecutive trials showing an inter-trial interval less than 1.1 s were
rejected, as well as trials with an EEG peak signal amplitude greater than 50 lVpp, since these trials would in all
probability have been caused by eye blinks. To compare the VEP waveforms between the two systems, signal
averaging was performed for an identical number of trials with a minimum of 32 trials per system for each
subject. We evaluated FVEP from the standard lead: Oz was referred to the left ear.

The reference and test EEG systems were compared serially with a 30 min break in between the recording
sessions. For the reference EEG system, the marked recording positions were prepared by the operator using a
standard EEG preparation paste, i.e., the wet electrodes were gelled, and secured in position using a bandage
glued with collodion. The contact impedance was kept below 5 k X. Signal evaluation was performed with regard
to looking for a peak in the evoked average potential at a standard time lag and amplitude for the P100
component (Lee et al., 2005; Takei et al., 1993).

3. Results

3.1. Reactive alpha rhythm

Experiment 1

While preparing the subject for the first two experiments, the preparation time per electrode was recorded and
averaged for all of the subjects. Based on our measurements the average set-up time for the reference system
using standard wet electrodes is within the range of 23 min per electrode and includes: spot preparation using
specialized abrasive paste, adding gel to the electrodes, electrode placement, sticking the electrode with
bandage, and checking of the contact impedance (and eventual re-placement in case the impedance is too
large). The average set-up time per dry electrode is about 10 s, which is mostly taken up by the time it takes
the collodion to dry. Fig. 4 shows a photograph of the electrode montage. In this figure, the dry electrodes are
indicated by the arrows and it is possible to observe that they are directly in contact with the scalp through a
thin film of collodion, while the wet electrodes are covered by white bandages.

Recordings of the alpha rhythm were made using both the new EEG system and the reference EEG system.
Visual inspection was used to ensure qualitative match between the two systems, before continuing with
analytical comparisons. Fig. 5 shows a recording in the time domain (10 s) of the signals recorded at the RPA
hospital from one subject using the electrode montage depicted in Fig. 3. In Fig. 5, dry electrodes signals are
represented in bold. The phenomenon is more apparent in the frequency domain as shown (Fig. 6). Observe the
difference in the spectrum around 9 Hz between the eyes open (bold) and eyes closed cases as recorded by a
dry electrode (solid line) and a standard wet electrode (dashed line) recorded in standard position C4 and Cp4,
respectively. Visually, both the dry and wet electrodes yield similar results and the data shown are
representative of the recordings across the eight subjects. As the test and reference EEG signals were recorded
on different computers, the signals were time-aligned by selecting the time alignment that maximizes the
cross-correlation of the two signals. Analytical comparisons, after appropriate time alignment of the signals,
showed that the average correlation coefficient between the dry electrode signal and the average of the four
surrounding standard electrode signals was 0.82 (across all the subjects).

Recordings were taken during a BCI task with both the test EEG system and the reference EEG system. Fig. 7
shows an example of a lateralized mu-rhythm pattern recorded from one subject while performing the imagined
BCI task. The phenomenon presents characteristic saw-tooth waveforms which are desynchronized across
electrodes. The average correlation coefficient between signals was calculated for three minutes of trial data
using a moving time window of one second. Fig. 8 shows an example of fine time alignment for one of the
signal workspaces recorded from one of the subjects. In this example it is possible to observe that the time
alignment between the two EEG machines differs by about 50 samples (maximum of the correlation index) from
the one calculated taking into account only the time stamp. The average correlation coefficient between signals
recorded with the same machine was 0.90 (for both dry and wet electrodes). The calculated average correlation
coefficient between each dry electrode and the average of the wet surrounding electrodes was 0.76. This value
is influenced by the presence of noise artifacts such as the one shown in Fig. 9. These artifacts are processed
differently in the two hardware systems due to the different recovery time and filter responses for each system
(Table 1). In further analyses, the signals were visually inspected by a blinded expert neurologist (AM). The
expert indicated periods within the trial that contained artifacts such as those shown in Fig. 9 and these
sections were removed from the data. The effect of the different filter response on the acquired signal is seen
quite clearly in the artifact pattern. When the correlation analysis was performed processing only the clean
signals, the best correlation between the dry electrodes and the average of the surrounding electrodes was
0.94 with an average of 0.83. Fig. 10 shows a comparison of the power spectrum for a wet and a dry electrode
signal after removal of artifacts (about two and half minutes of data).
3.4. FVEP Experiment 3

In this experiment the dry electrode was held in position using an elastic bandage. Because this configuration
was uncomfortable for the subject due to the strong compression on the head and in order to minimize the
error due to the electrodes placement across the trials, we decided to evaluate the unipolar FVEP left ear
referred rather than the FVEP referred to Fz (as recommended in the standard procedure). Fig. 11 shows the
FVEP signals recorded by the two systems. Experiments across three subjects show that for the same subject, a
variation of less than 10% in amplitude of P100 was obtained for the two EEG systems. This variation is due to a
number of parameters such as variance in gain due to component tolerance, subject adaptation, slight
differences in hardware filter responses and an imprecise match between electrodes area and position.

4. Discussion

The three field experiments demonstrate that the new EEG system is comparable with reference EEG systems
while requiring substantial less set-up time. Moreover, while we have examined the new EEG system in a
number of arbitrary test scenarios, it is important to clarify the possible use cases in which the new EEG system
may offer advantages.

Dry electrodes secured in position with collodion offer improved reliability and robustness for long-term
monitoring. While dry electrodes could be held in place using different approaches such as pressure as
described in Popescu et al. (2007), we have found that most designs that keep the electrode pressed against
the scalp are uncomfortable for long-term use (e.g., in the FVEP evaluation). Therefore, the dry electrodes
presented here have been designed to be glued in position using just collodion. Collodion is basically a non-
conductive and skin compatible super glue. Our laboratory tests on commercially-available collodion samples
show that thin films of collodion applied to the terminals of a 1 k X resistor increase its ohmic value to over the
maximum readable value of 20 MX associated with our multimeter. Early studies, such as one presented by
Blinks (1930), confirm that the addition of a collodion film adds a random electrical impedance up to several
MX. As described in standard clinical applications (Tyner et al., 1983), collodion is commonly used to secure a
bandage that keeps the electrode in position. Instead, in the new EEG system, a drop of collodion can be
directly applied on the flat surface of the dry electrode. In order to improve the grip further, collodion could be
applied to the boundary of the electrode once it is in position. In the case of bristly hair, we found that the hair
should be spread using two fingers to avoid sticking the electrode to the hair rather than the scalp. In addition,
preventive hair de-greasing with an alcohol swab seems to facilitate the collodion grip and may be used in
situations in which hair-dressing products pose a problem. It is important to note that passive dry electrodes
which do not contain any electronics are cheap to manufacture and are designed to be disposable. Moreover
our prototype of passive dry electrode (pictured in Fig. 12 together with a standard golden brass EEG
electrode), in Fig. 2, was made with medical grade silicone rubber and in the event of being re-used, it is
compatible with all the known sterilization methodologies for this materials.

However, there are difficulties and disadvantages with this method. For example, we have observed that the
absence of con- ductive gel or paste between the scalp and the electrode makes accentuates artifacts
associated with facial movements (such as grinding teeth) or subject movement. An example of this is visible in
Fig. 9, where the signal from C3 shows a low frequency deviation that radically changes the shape of the signal
during the artifact and a substantial the recovery time after the artifact. The motivation for developing new
hardware for EEG recording stems from the need to reduce the discomfort for patients and the time required for
preparation. Ultimately we hope that it will lead to a more reliable and flexible portable system for long-term
monitoring. In this regard, the main defining characteristic of the new EEG system is that it is compatible with
passive dry electrodes. In contrast to dry electrodes, we first briefly discuss the use of ready-made elastic caps
with embedded tin or silver/silver chloride electrodes which offer potential long-term monitoring with the
advantage of eliminating the head measurement and spot identification task. It is important to emphasize that
we see dry electrodes as compatible with ready-made caps, but not necessarily reliant on them. A primary
disadvantage with the elastic cap is its rigid structure. In other words the caps are available in standard sizes
such as XS, S, M and L and this means that the cap stretches differently for various head shapes and sizes.
Adjustable caps may also be used, but generally in this case the electrode spots are identified before-hand.
With regard to long-term monitoring, elastic caps may suffer a loss of tension, or worse, may slide off position
due to head movement.

5. Conclusion We have presented a new EEG system that is compatible with passive dry electrodes. Given the
results obtained in three field of experiments conducted to compare the new system with various different
reference EEG systems: alpha rhythm, mu-rhythm signals that were highly correlated with signals recorded in
parallel with a clinically-approved system and the FVEPs patterns which had differences smaller that 10% in the
amplitude of the P100 feature, which is considered as a marginal difference in clinical applications. We believe
that passive dry electrodes may offer advantages for EEG monitoring (e.g., ease of application, stability and
reliability). Generally, the time required to apply collodion to the dry electrodes (10 s per electrode) is much
less than the time required to prepare wet electrodes, while the time that the dry electrodes remain viable once
attached is longer than that for wet electrodes. One of the challenges we see remaining and we will focus on in
the future is how to hold the dry electrodes in position without requiring collodion (or any other type of glue)
while still remaining suitable for long-term monitoring. In summary, we believe that the new EEG system moves
the state of the art closer towards one of the holy grails for EEG recording system, namely that it be (1) easy to
use, requiring little time to install and remove; (2) portable, so people can use it at home and during daily
activities; and (3) wireless, so in applications such as BCI the data can be processed remotely by a more
powerful computer. Acknowledgments The Penso project was supported by a University of Sydney research
grant. Subjects consented to the study approved by the University of Sydney Human Ethics Committee and the
University of Naples correspondingly. Authors gratefully acknowledge the contributions of RPAH staff. The
authors thank Burr-Brown Semiconductor, Maxim semiconductor and Samtec, for kindly providing samples.