Detection of Electroencephalography Artefacts using Low

Fidelity Equipment

Patrick Schembri, Richard Anthony and Mariusz Pelc

Department of Computer and Information Systems, University of Greenwich, Greenwich London, U.K.

Keywords: Artefacts, Classification, Brain Computer Interface, EEG, Electroencephalography.

Abstract: The use of Electroencephalography (EEG) signals in the field of Brain Computer Interface (BCI) has gained

prominence over the past decade, with the availability of diverse applications especially in the clinical

sector. The major downside is that the current equipment being used at medical level is specialized,

complex and very expensive. Our research goals are to further increase accessibility to this technology by

providing a unique approach in data analysis techniques, which in return will allow the usage of cheaper and

simpler EEG hardware devices targeted for end users. We use non-invasive BCIs designed on EEG, mainly

due to its fine temporal resolution, portability and ease of use. The main shortcoming of EEG is that it is

frequently contaminated by various artefacts. In this paper we provide vital groundwork by identifying and

categorizing artefacts using low fidelity equipment. This work forms part of a wider project in which we

attempt to use those artefacts constructively, when others try to filter them out. The main contribution is to

create awareness of the extent to which artefacts can be encountered, identified and categorized using off-

the shelf equipment. Our results illustrate that we are able to adequately identify and categorize the most

commonly encountered artefacts in a non-clinical environment, using low fidelity equipment.

1 INTRODUCTION

This paper discusses the artefacts of a non-invasive

BCI (Brain Computer Interface) on the basis of EEG

(Electroencephalography) where the signals will be

extracted from the electromagnetic (EM) brain

functions without the use of muscular activity.

Initially EEG was targeted for use in clinical

applications with patients that have medical

conditions such as Lou Gehrigs disease (Allison et

al., 2012). However over the past decade the use of

biomedical signals has also increased significantly in

non-clinical applications. This has led to the

development of a number of devices that can be

controlled by signals emitted from the brain.

At the present time, human BCI research has

been developing into two main areas; invasive and

non-invasive. The most prevalent invasive

techniques are called Electrocorticography (ECoG)

or intracortical recordings, which have their

electrodes in direct contact with the cerebral cortex

while the most prevalent non-invasive technique is

called Electroencephalography (EEG) which has its

electrodes placed along the scalp surface (Dornhege

et al., 2007). The qualitative difference between

these areas is that invasive BCI has a much better

signal quality with higher amplitudes and spatial

resolutions; it has a high signal-to-noise ratio and is

less susceptible to artefacts; however it requires a

surgical intervention for electrode placement. On the

other hand non-invasive BCI has a much weaker

signal and is prone to a number of different artefacts.

However it has an excellent temporal resolution

(Ball et al., 2009) and does not require any surgery.

In addition to using non-invasive BCI based on

EEG, our research also makes use of low cost off the

shelf equipment. The aim is to increase accessibility

to this technology by providing a unique approach in

data analysis techniques, which in return will allow

the usage of cheaper and simpler EEG hardware

devices targeted for end users. This paper does not

imply that the low fidelity equipment being used

could replace medical equipment; as a matter of fact

it does not have any certification; therefore it should

be employed sensitively for non-clinical trials.

(Frey, 2016) states that “Open-hardware initiative

does not aim at medical applications, hence it

should be employed in sensitive contexts.”

An artefact is a signal that is detected by EEG

equipment, which is not of cerebral origin but from

Schembri, P., Anthony, R. and Pelc, M.

Detection of Electroencephalography Artefacts using Low Fidelity Equipment.

DOI: 10.5220/0006398500650075

In Proceedings of the 4th International Conference on Physiological Computing Systems (PhyCS 2017), pages 65-75

ISBN: 978-989-758-268-4

various different sources. In the context of EEG,

artefacts are unwanted since they mask the brain

wave signals; however they could potentially be

used as a primary interface. According to the

glossary of the International Federation of Clinical

Neurophysiology (IFCN), the term artefact is

described as “A potential difference due to an

extracerebral source, recorded in EEG tracings”;

which is expanded to “A modification of the EEG

caused by extracerebral factors such as alterations

of the media surrounding the brain, instrumental

distortion or malfunction, and operational errors”

(Noachtar et al., 1999). Artefacts have always been

given great importance in the context of EEG due to

the undesirable affect that these have on the signal of

cerebral origin.

The work presented here is part of a larger EEG-

based project, and thus it is important to recognise

and understand the artefacts that are detectable.

These artefacts are usually an unwanted signal in the

context of EEG; however we are interested in using

them as part of a control interface. In this paper we

prepare the groundwork for filtering and using these

artefacts, through categorization of artefacts, and

their manifest characteristics, using specific low

fidelity equipment.

2 EMPIRICAL INVESTIGATION

OF EEG ARTEFACTS USING

OFF-THE-SHELF EQUIPMENT

Our work is concerned with exploring the

capabilities and limitations of low cost off the shelf

equipment which in return will facilitate and

increase accessibility for EEG applications. We aim

to compensate for the low fidelity aspect of this

equipment with enhanced software filtering and

analysis. This particular part of the work sets the

foundations for further work by investigating the

way in which various artefacts are detected,

identified and categorized with low fidelity

equipment.

A way in which an electrode (input1) is

connected relative to another electrode (input2) is

called a derivation. A collection of derivations are

called a montage and there are several different ones

in popular use. The intention of using a specific

montage is to keep the experiments tractable and to

avoid unnecessary complexity. Moreover other

types of montage; even the more complex such as

Laplacian and Common Average Reference; can be

derived from the collected data, since montage

reformatting is achieved by performing a simple

mathematical operation. In fact (Fisch, 1999) states

that “for this reason, digital EEG systems store the

original EEG signal in a referential montage

containing all electrodes”. This is of course possible

as long as all the electrodes that need to be

combined have in some way been referred to each

other in the original recording.

For instance when labelling a channel montage

as Fp1-A1, a mathematical expression is being

created which implies that the signal displayed will

be Fp1 minus A1. If a recording has been obtained

from Fp1-A1 and Fp2-A1 then Fp1-Fp2 can be

derived from:

(Fp1 - A1) - (Fp2 – A1) =

Fp1 - Fp2 + A1 - A1 = Fp1 - Fp2

(1)

Although montage reformatting is possible to be

performed instantaneously, this is ideally used for

recorded sessions and is not suggested for real-time

streaming.

2.1 Equipment Used

The work reported herein is based on an OpenBCI

32-bit board connected with an Electro-Cap

using

the international 10/20 system for scalp electrode

placement in the context of EEG experiments. A

basic overview of the equipment being used is

shown in Figure 1.

Figure 1: OpenBCI Board and Electro-CAP.

The OpenBCI 32-bit’s board microcontroller is the

PIC32MX250F128B

which includes a 32-bit

processor with a maximum speed of 50MHz; storage

of 32KB of memory and is Arduino compatible.

The board uses the ADS1299

IC developed by

Texas Instruments, which is an 8-Channel, 24-Bit,

simultaneous sampling delta-sigma, Analogue-to-

http://openbci.com/

http://electro-cap.com/

http://www.microchip.com/wwwproducts/en/en557425

http://www.ti.com/product/ADS1299

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

Digital Converter used for bio potential

measurements such as in EEG and

Electrocardiography (ECG). The 24-bit resolution

gives a huge range of microvolts (µV) that covers +-

187mV (187000µV); the working is shown in

section 2.2. When considering that EEG data ranges

are typically between +-100µV, it illustrates that it is

able to provide a broad spectrum of flexibility and

scalability. Moreover this chip is capable of

supporting up to 16,000Hz although the transfer of

that much data through an Arduino would be

impracticable. There is the ability to use the SD card

for faster sample rates, which is discussed below.

The board comes with eight bio potential input

channels which can be increased to sixteen channels

with the addition of a Daisy Module; which plugs

itself onto the existing OpenBCI 32-bit board. Our

current experiments do not make use of the daisy

module, although future experiments may need these

extra channels.

The system comes with a pre-programmed USB

dongle for wireless communication which

communicates with the low cost RFDuino

RFD22301 microcontroller built on the OpenBCI

board. This microcontroller can communicate

wirelessly with any device compatible with

Bluetooth Low Energy (BLE). In addition a local

Secure Digital (SD) slot is built-in the board, which

gives it the ability to store recorded data on SD

memory card. This is particularly useful when

requiring improved portability and highest data

rates.

An additional feature which is included in the

OpenBCI board is a 3-axes accelerometer from ST

with model LIS3DH

. This accelerometer is capable

of 16 bit data output and of measuring accelerations

with output data rates from 1 Hz to 5.3 kHz. This

can prove to be quite useful; such as, for sensing

change in orientation of the head or sensing rough

motion. In these cases the value from the

accelerometer would suggest that motion artefacts

would be within the EEG data. In our experiments

this information was not required, since the board

was firmly placed on the desk. However in the

future we are planning on using the OpenBCI

Ultracortex MK4 cap, which has the ability of

attaching the board to the actual headset, where the

data from the accelerometer would be extremely

valuable. Figure 2 depicts a graphical representation

of these components.

http://www.rfduino.com/

http://www.st.com/en/mems-and-sensors/lis3dh.html

Figure 2: OpenBCI Board Components.

The Electro-Cap being used in our experiments has

the fabric which is made from elastic spandex and

has recessed pure tin wet electrodes directly attached

to the fabric. The term wet electrodes type, implies

that the use of an electrolyte gel is required to make

effective contact with the scalp; otherwise it may

result in impedance instability.

2.2 Experimental Setup

The EEG signals where sampled at 250Hz (this

being OpenBCIs default value) while the sampling

precision was 24-bit. The recordings where stored

anonymously as raw data in text, comma separated

value (csv) files. Eight EEG electrodes where used

in different regions of the scalp according to the

International 10-20 System as shown in Figure 3.

This system is the de facto standard for the

placement of electrodes along the head. Each

electrode is assigned a letter to identify the lobe and

a number to identify the hemispheric location. The

letters F, P, T and O stand for Frontal, Parietal,

Temporal and Occipital lobes. In addition, letter C

refers to the central area of the brain. The even

numbers represent the electrodes positioning on the

right hemisphere, while the odd numbers represent

the electrodes positioning on the left hemisphere.

The Xz stands for a zero and represents an electrode

placed on the midline such as Fz, Cz and Pz. In

addition the letter A can represent the reference

electrode which will measure the potential

difference between itself and the other electrodes

and/or the ground electrode for common mode

rejection.

The equipment we are using supports the use of

eight electrodes. The electrode positions Fp1, Fp2,

C3, C4, T5, T6, O1 and O2 are selected because

they provide good coverage for detecting these

artefacts. These are referenced to the electrode A1 as

follows: Channel 1: Fp1; Channel 2: Fp2; Channel

3: C3; Channel 4: C4; Channel 5: T5; Channel 6:

T6; Channel 7: O1; Channel 8: O2 as shown in

Figure 3. A referential montage was selected to

Detection of Electroencephalography Artefacts using Low Fidelity Equipment

analyse how artefacts are exposed with this setup,

even though no single reference electrode is ideal for

all situations. Nonetheless and if required, other

types of montage can be reconstructed from the

chosen montage by executing a simple mathematical

operation; as previously explained. The reference

electrode was placed on the left earlobe A1 as shown

in Figure 4.

Figure 3: Electrode placement following the International

10-20 system.

EEG signals where obtained from a healthy human

subject; male in the age group between 30 and 40

years old and on three different sessions with a few

days apart. Before the start of the experiments, the

subject was asked to calm down in a seated position

and relax for a few minutes. The subject was seated

one meter away from the equipment. The researcher

and his equipment where situated on the left side of

the subject. Then, the subject was instructed on a

series of tests such as muscle movement that are

designed to detect the artefacts which are discussed

in Section 3.

Three trials where conducted for these

experiments. The first session results and recordings

where archived. The second session was done on a

separate day with the same conditions of the first

session and the results where archived for

comparisons. These two sessions were carried out to

familiarize the user with the equipment and the

methodology of the experiments. The third session

was done a day later with the same conditions of the

first and second session and the results are shown in

this paper. During the recording the subject received

a 2 second beep sound to perform the requested trial

and a 1 second beep sound to stop.

Figure 4: Referential Montage used.

2.3 Processing

The data that was transmitted from the RFDuino

module found on the OpenBCI board is considered

as ‘raw’ EEG data in ADC counts. These where

transferred as 24-bit integer, since it’s the native

format used by the ADS1299 chip. Since this is an

unusual format, it was immediately converted via

the OpenBCI open-source JAVA function

‘interpret24bitAsInt32’ into a 32-bit signed integer

(Audette, 2014).

Subsequently the scale factor was required,

which is the multiplier used to convert the EEG

values from counts to scientific units like volts. This

is found by following the formula in the ADS1299

datasheet table number 7:

Scale Factor = V

REF

/ (2

-1) / Gain * 1000000

(2)

The datasheet also states that the voltage

reference input is hardware bound to 4.5volts, while

we used the maximum and default gain factor of 24-

bit. Thus the formula (2) can be reformed into:

Scale Factor = 4.5v / (2

-1) / 24 * 1000000

(3)

Hence the scale factor value is 0.02235 per

count. Therefore the 32-bit signed integer is

multiplied by the scale factor and we get the EEG

data values in microvolts (µV). This is the actual

stored data in the csv file. The full scale of +-

187mV (187000µV) discussed in Section 2.1 is

achieved by 2

* 0.02235 = +-187485.388µV.

As previously mentioned the ADS1299 chip is

capable of a sample rate of up to 16,000Hz; however

in our experiments we used OpenBCIs default rate

of 250Hz especially when considering that the data

was being transmitted wirelessly through the

RFDuino module.

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

The captured raw data was imported in

MATLAB R2014a

via the csvread command into a

MATLAB matrix and any unnecessary rows and

columns were removed. These consisted of the first

five rows which are superfluous comments; the first

column which stored the sample index / packet

counter and the last three columns which stored the

auxiliary data of the accelerometer.

The MATLAB array was later imported into

EEGLAB

for processing and for offline qualitative

and quantitative analysis. The first process was to

apply a 50Hz (60Hz in some countries) notch filter

to eliminate the environmental electrical

interference, which was only omitted for the

50/60Hz artefact experiment. In addition a high pass

filter was applied at 0.5Hz to remove the DC offset

and a low pass filter of 49Hz was applied to remove

any signal harmonics and unnecessary frequencies

which are not beneficial in our experiments. As an

alternative a band-pass filter of 0.5Hz-49Hz could

have been chosen, however it was not selected since

this type of filter does not attenuate all frequencies

outside the range. In fact the filter’s frequency

response function is not very steep; it doesn’t

completely cut-off at the required frequency, but

instead it rolls off more gently with the frequency.

The result from this processing yields a rich EEG

signal for our experiments which can be analysed

with different tools. The screenshots presenting the

EEG signal (see Figures 5-15) where plotted by

using the EEGLAB Plot: Channel Data (Scroll)

menu option. The frequency-time domain

screenshots where produced by the Time-Frequency

transforms: Channel-time frequency menu option.

The plot Event Related Spectral Power (ERSP) was

employed since it is a statistical measure; the mean

of a distribution of single-trial time/frequency

transform (Neuper & Klimesch, 2006). In our

processing we used the Fast Fourier Transform

(FFT) option; 400 time points for the time-frequency

decomposition and the frequency was set between

one and forty which provides us with enough

information for artefacts detection. The baseline was

set to the default of 0 for pre-stimulus and the single

trial DIV baseline option was used. Subsequently the

choice of channel number and time range in relation

to the experiment being analysed where entered

(such as Channel 1 for FP1; time range 5000ms –

9000ms).

The spectrogram frequency-domain screenshots

were produced in Matlab; outside of EEGLAB. The

https://www.mathworks.com/products/matlab.html

https://sccn.ucsd.edu/eeglab/

data was filtered using Butterworth filter design of

the second order. First a notch filter was used

followed by a low pass and a high pass filter; with

the same values used for EEGLAB. The actual code

for the filtering and the spectrogram are shown in

the appendix section.

3 ARTEFACTS - RESULTS

Although a number of research papers have been

published showing different types of artefacts such

as (AYDEMIR et al., 2012) and (Begum, 2014);

these were presented with a “black box” approach or

using medical equipment, or otherwise, mentioned

in a different context. What we present in this paper

are results that are relevant to our own specific low

fidelity hardware.

An EEG device is very sensitive and it is easily

susceptible to disruption from other electrical

activities. Moreover some artefacts are easily

distinguishable while others closely resemble

cerebral activity and are very challenging to be

recognized. Artefacts are usually categorized as

physiological (biological) and non-physiological

(extra physiological) (Fisch, MD, 2000). The

classification mentioned below is not rigorous; for

instance, if the subject makes a movement, this may

lead to artefacts originating as electrode artefact.

Even though signal artefacts caused by non-brain

wave signals can be problematic when studying

brain waves directly, the signal artefacts themselves

could be used directly as command signals within an

interface.

3.1 Physiological Artefacts

Physiological artefacts are bioelectrical signals that

are generated from the user’s body excluding the

brain. These are usually embedded along the

electrical cerebral bio-signals in an EEG session.

The physiological artefacts include, but are not

limited to:

3.1.1 Ocular Artefacts

Ocular artefacts are essentially a result from the

eyeball acting as a dipole which becomes pertinent

when it develops into a moving electrical field such

as when the subject opens and closes his eyes and/or

the EMG potentials from muscles in and around the

orbit. These generate signals that are detected

predominantly by electrodes Fp1/Fp2 and F7/F8.

Detection of Electroencephalography Artefacts using Low Fidelity Equipment

1) Blink: blink/blinking which is the most

common ocular artefact, occurs spontaneously and is

very challenging for the subject to control even for

short periods of time. When the subject blinks, the

eyeball triggers an instinctive upward movement

(Bells phenomenon) and hence produces a positive

potential in the frontal lobe which is displayed in

EEG as a transient, diphasic, synchronous slow

wave (Misra & Kalita, 2005) (Stern, 2005)

(Sovierzoski et al., 2008) as shown in Figure 5. This

image also shows that, the faster the blink the

shorter the wavelength, as depicted by the first blink

occurrence which was faster than the second blink.

When the subject performs a number of repetitive

blinks, the displayed EEG could mimic a triphasic

wave or resemble rhythmic delta activity as shown

in Figure 6. Additional and more frequent blinking

can simulate theta activity as shown in Figure 7.

Figure 5: Ocular Artefact - Eye Blink predominantly on

Electrodes Fp1 and Fp2 (Plot, ERSP, and Spectrogram).

Figure 6: Ocular Artefact – Repetitive Eye Blinks (Plot,

ERSP, and Spectrogram).

Figure 7: Ocular Artefact – More Repetitive Eye Blinks

(Plot, ERSP, and Spectrogram).

2) Eye Flutter: Eye Flutter produces an ocular

artefact that is more rhythmic, with higher frequency

and lower amplitude as shown in Figure 8.

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

Figure 8: Ocular Artefact – EyeFlutter (Plot, ERSP, and

Spectrogram).

3) Lateral Eye Movement: Lateral Eye

movement artefact is mostly detected in a bipolar

longitudinal montage using Fp1-F7 and Fp2-F8 and

may start off with a single sharp muscle potential

called lateral rectus spike. In this type of montage a

left lateral eye movement will have a positive

potential in electrode F7 and an opposite negative

potential in electrode F8. In our referential montage,

the frontal origin of eye movement artefacts

remained indistinguishable due to the reference

electrode (A1) being contaminated by eye

movements (Fisch, MD, 2000).

4) Slow Roving Eye Movement: Slow Roving

eye movement differs from lateral eye movement

since no saccades occur; consequently resulting in

no abrupt changes. On a bipolar montage these are

reflected as smooth lateral movements with phase

reversing. On a referential montage using low

fidelity equipment, this artefact was not detected.

3.1.2 Muscle Artefacts - EMG

(Electromyography) Activity

EMG activity produces artefacts that are due to

muscle contraction and are the most common and

significant noise source in the context of EEG.

Although EMG in itself is useful for

electromyography; they are considered noise in

EEG, since they overlap and obscure the EEG signal

due to their higher amplitude and frequency. If,

however, this signal is passed through a low-pass

filter set at 35Hz or less, this will change their form

and caution is required since these may transpire as

beta activity or like abnormal epileptiform spikes.

The extent of a muscle artefact depends on the

duration of the muscle activity, which might be less

than a second and/or throughout the entire session

(Stern, 2005) (Fisch, MD, 2000) (Misra & Kalita,

2005).

1) Surface EMG: Surface EMG activities

generally occur in regions with underlying muscle

such as the masseter and temporalis muscle, which

affect the frontal and temporal electrodes. These

may also disseminate and diffuse to other channels.

Electrodes Fz, Cz and Pz can provide a reasonably

pure EEG signal. Figure 9 shows an EMG effect

when the subject clenches his teeth. ERSP

screenshot doesn’t show any recognizable activity.

Figure 9: Muscle Artefact – Clench Teeth (Plot, ERSP,

and Spectrogram).

2) Glossokinetic: Glossokinetic is an artefact

arising from the movement of the tongue. It is

similar to the eyeball movement in ocular artefacts,

though less sharp. The tongue functions as a dipole

where the tip acts as a negative with respect to the

positive base. This results in the surging of diffuse

delta like activity, which is frequently supplemented

by muscle artefact. The tongue has a DC potential

and equipment running on DC amplifiers will not

record its potential as is the case in the equipment

being used for this experiment.

Figure 10 shows the effect of swallowing in our

subject which affects the oropharyngeal muscle.

This experiment could have been included in the

Surface EMG section, since no tongue potential is

being recorded, but is being listed here for

classification reasons.

Detection of Electroencephalography Artefacts using Low Fidelity Equipment

3) Intermittent Photic Stimulation (IPS):

Intermittent Photic Stimulation (IPS) / is a

photomyogenic / photomyoclonic response to a

visual stimulation where the subject eyes are

presented with intermittent flashes of light. This

results in an involuntary time linked facial muscle

response to the flash of light which affects the

frontal and periorbital regions, specifically the

frontalis and orbicularis muscles (Shamsaei, n.d.).

At this stage in our work we don’t include this.

Figure 10: Muscle Artefact – Swallow (Plot, ERSP, and

Spectrogram).

3.1.3 Movement Artefacts

Movement during an EEG session may produce two

distinct artefacts; instrumental from the movement

effect on the electrodes and their leads as discussed

in the Equipment Artefacts section below; and

biological through the generation of electrical fields

from muscle contraction; EMG activity; as discussed

in the Muscle Artefacts section above.

3.1.4 Cardiac Artefacts

Electrocardiography (ECG) is the process of

recording electrical activity from the heart. The heart

produces a considerable electrical field that spreads

to the base of the skull, which is detectable in an

EEG session. This artefact is easily detected in a

referential montage since there is ample

interelectrode distance between the reference which

is located on the ear lobe and the other electrodes

which are located on the scalp. In addition this

artefact is most prominent in subjects with a short

neck. This artefact appears as a QRS complex which

represents three graphical deflections in an ECG

diagram. The QRS complex is preceded by a P wave

and followed by a T wave as shown in Figure 11.

With clinical EEG equipment using a referential

montage setup; a poor QRS complex was formed.

This was due to the distance from the heart where

the P wave and T wave are not visible (Fisch, MD,

2000) (Stern, 2005). ECG artefact may be reduced

or removed by adding a second reference; however

it will only work if both reference electrodes are able

to detect a pulse (Spriggs, 2010). Unfortunately we

were unable to reproduce this artefact using low

fidelity equipment. It is true that the artefact is a

poorly formed QRS complex which is most

prominent in short necks and could have been easily

concealed within the noise; but that does not negate

the fact that we should have at least encountered it

even as a low amplitude signal. We have tried

several types of filters but without any apparent

result. We were however able to produce an ECG

signal on purpose; not as an artefact; with a different

set-up, which however is beyond the scope of this

paper.

Figure 11: QRS Complex.

3.1.5 Pulse Wave Artefact

Pulse artefact mainly occurs when electrodes are

placed over a pulsating artery manifesting a regular

pulse beat. These pulsations instigate periodic slow

waves that can be misidentified as EEG activity.

There is a direct link between ECG and pulse waves;

where the QRS complex happens right before (about

200ms) the pulse waves. In our experiments the

electrodes where not placed over a pulsating artery

and thus it did not show in our experiments.

3.1.6 Skin Potential

Skin potentials where discussed in Non-

Physiological Artefacts, explicitly under Equipment

Artefacts which included Perspiration and Salt

bridges.

3.2 Non-Physiological Artefacts

Non-Physiological artefacts are externally generated

outside the user’s body such as artefacts arising from

environmental electrical interference and artefacts

relating to the equipment being used.

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

3.2.1 Environmental Electrical Interference

Environmental Electrical Interference: 50/60Hz

Artefact; The most common electrical interference

artefacts usually emanate from electrical devices and

in close proximity to power lines. The greatest

contributor is the alternating current (AC) with a

monomorphic frequency of either 60Hz (ex. United

States) or 50Hz (ex. Europe). These artefacts can be

introduced either electromagnetically, where the

strength of the field is determined by the current

flowing through cables or by the equipment such as

transformers and TV power supplies; and

electrostatically due to the capacitance property of

objects where the subject or electrodes pick up

capacitance potentials from other sources which are

in their proximity such as the movement of any

charged bodies or objects (ex. plastic, rubber,

synthetic fibres) near the subject (Fisch, MD, 2000)

(Binnie et al., 1982). Figure 12 shows the effect of a

50Hz noise on our EEG signal.

This artefact can be reduced by grounding the

equipment, moving the subject away from power

lines and sources that can generate electrostatic

interference and keeping electrodes impedance to

less than 5KΩ which is the leading cause of the

50/60Hz artefact (Spriggs, 2010). Should these

methods not suffice; the artefact can be eliminated

by a notch-filter (or similar) which will only remove

the 50Hz or 60Hz activity from the signal. The filter

should only be used if necessary.

Radio Frequency / Mains-Borne: Other electrical

interferences which are less prominent include

Radio Frequency when they are modulated in a

lower frequency and Mains-Borne interference

arising from fluctuating power supplies.

Figure 12: Electrical Artefact: 50Hz (Plot, ERSP, and

Spectrogram).

3.2.2 Equipment Artefacts

A number of different artefacts can be caused from

the recording electrodes and the equipment being

used. Electrode artefacts can manifest as two

dissimilar waveforms; low frequency rhythms

amidst a scalp area and brief transient morphology

which would be limited to one electrode (Stern,

2005).

Electrode Pop: Electrode Pop can occur

occasionally when there is an instantaneous change

in the electrical potential between the electrode and

the scalp, where it is typically followed by a sudden,

high amplitude spike in the EEG recording (Barlow,

1986) as shown in Figure 13. This may occur when

electrodes are not firmly attached and/or when direct

pressure is applied on the electrodes.

Figure 13: Equipment Artefact: Electrode Pop (Plot,

ERSP, and Spectrogram).

Electrode Contact and Lead Movements: A weak

Electrode Contact and Lead Movements generate a

different artefact that has a less sustained

morphology compared to electrode pop as shown in

Figure 14. The weak electrode contact results in

impedance instability, which will produce waves

with fluctuating amplitude and morphology;

although if there is a context of rhythmic movement

such as from tremors, the resulting waves may be

rhythmic as well. Lead movements do not resemble

any true EEG activity where the morphology of the

wave is incoherent (Stern, 2005).

Salt Bridge: A Salt Bridge artefact can occur

when smearing the electrolyte gel between two

electrodes or by applying an excessive amount of

electrolyte gel, which may result in an inadvertently

overlap, thus creating a short circuit between the

electrodes. This artefact is usually channel specific

and manifested as a low amplitude wave compared

to the background. Salt bridge artefact will

Detection of Electroencephalography Artefacts using Low Fidelity Equipment

eventually be prevented by use of dry electrodes;

which we plan to do in our future experiments.

Perspiration: Perspiration artefact although not

as stable, is similar to a salt bridge artefact where the

salinity between electrode locations will merge the

affected electrodes as a single entity. It is usually

manifested as a slow wave that is typically greater

than 2 seconds in duration which is out of the

frequency scope of EEG (Stern, 2005) (Fisch, MD,

2000).

Figure 14: Equipment Artefact: Electrode Contact and

Lead Movement (Plot, ERSP, and Spectrogram).

Salt Bridge and Perspiration artefacts can be easily

recognized in an EEG session and should be

resolved prior to commencement. The salt bridge

artefact is eliminated by cleaning the excess

electrolyte between the affected electrodes and

wiping the subjects forehead with a spirit solution,

while the perspiration artefact can be eliminated by

providing a cooler environment and reducing the

emotional stress of the subject. The experiments

reported here where based on a referential montage,

where these artefacts where not present. The lack of

these findings suggests that an electrolyte bridge is

only present amongst electrodes such as in a bipolar

montage.

4 CONCLUSION

Non-invasive BCIs designed on EEG provides fine

temporal resolution, portability and ease of use

however the signal is frequently contaminated by

various artefacts. EEG processing and analysis

require accurate information and it is vital that these

artefacts are recognized and classified so that it

would be possible to eliminate or prevent them from

occurring, or otherwise, attempt to use them

constructively.

Previous investigations in this research area

where made using expensive medical EEG

equipment and were usually categorized using

different type of montages, which made it

challenging for comparisons. Moreover only a few

of these artefacts have been documented

successfully using low fidelity equipment and this

documentation has been ad hoc and not categorized

properly.

Due to the proliferation of cheap EEG

equipment, including user-made equipment such as

(Wang et al., 2016), an evident necessity to validate

the equipment’s suitability was present. Moreover in

recent times, a number of researchers and end-users

are using low fidelity equipment as a “black box”

approach (Lecoutre et al., 2015), without any

qualitative testing on the equipment being used.

Part of our contribution was to create awareness

of what type of hardware components are being

utilized in low fidelity equipment, vis-à-vis the

results achieved. This would ultimately facilitate the

possibilities of using off-the-shelf EEG equipment

as a cheap alternative to medical EEG equipment. It

is important to note that this paper does not imply

that low fidelity equipment should replace medical

equipment; our purpose is to assess the suitability of

such equipment for non-clinical trials.

In this paper, a successful approach in identifying

and classifying artefacts using low fidelity

equipment on a referential montage is presented.

The promising results achieved show that the most

common artefacts observed in a non-clinical

environment are being effectively identified and

categorized while using the aforementioned

equipment.

5 FUTURE WORK

Future research work includes the capability of low

fidelity equipment, to accurately capture Mu and

Alpha waves/rhythms which can be processed to

perform tasks such as motor control functions. Some

initial results are shown in Figure 15, where the

Alpha waves are predominantly seen in the occipital

lobe, specifically on O1 and O2 electrodes, whereas

the Mu waves are predominantly found around the

central area of the brain known as “central sulcus”,

specifically on C3, Cz and C4 electrodes in our

figure. In addition we are also interested in exploring

the idea of using some of these artefacts

PhyCS 2017 - 4th International Conference on Physiological Computing Systems

constructively in concurrence with actual brain wave

signals.

Figure 15: Initial results for Alpha & Mu Waves.

REFERENCES

Allison, B.Z. et al., 2012. Towards Practical Brain-

Computer Interfaces. Springer.

Audette, C., 2014. Data Format for OpenBCI V3. [Online]

Available at: “https://github.com/OpenBCI/OpenBCI-

V2hardware-DEPRECATED/wiki/Data-Format-for-

OpenBCI-V3” [Accessed February 2017].

Aydemir, O., Pourzare, S. & Kayikcioglu, T., 2012.

Classifying Various EMG and EOG Artifacts in EEG

Signals., 2012.

Ball, T. et al., 2009. NeuroImage. Signal quality of

simultaneously recorded invasive and non-invasive

EEG.

Barlow, J.S., 1986. Automatic Elimination of Electrode-

Pop Artifacts in EEG's. IEEE Transactions on

Biomedical Engineering, Vol. BME-33, No. 5.

Begum, B.S., 2014. A Review on Machine Learning

Algorithms in Handling EEG Artifacts. The Swedish

AI Society (SAIS) Workshop SAIS, 14.

Binnie, C.D., Rowan, A.J. & Gutter, T.H., 1982. A

Manual of Electroencephalographic Technology.

Cambridge University Press.

Dornhege, G. et al., 2007. Toward Brain-Computer

Interfacing. Massachusetts Institute of Technology.

Fisch, MD, B., 2000. EEG Artifacts. LSU Medical Center.

Fisch, B.M., 1999. Basic Principles of Digital and Analog

EEG. Elsevier.

Frey, J., 2016. Comparison of an Open-hardware

Electroencephalography Amplifier with Medical

Grade Device in Brain-computer Interface

Applications. In PhyCS., 2016. SCITEPRESS –

Science and Technology Publications.

Lecoutre, L. et al., 2015. Evaluating EEG Measures as a

Workload Assessment in an Operational Video Game

Setup. In PhyCS., 2015. SCITEPRESS – Science and

Technology Publications.

Misra, U.K. & Kalita, J., 2005. Clinical

Electroencephalography. Elsevier.

Neuper, C. & Klimesch, W., 2006. Event-Related

Dynamics of Brain Oscillations. Elsevier.

Neuroscience, S.C.f.C., 2016. EEGLab. [Online]

Available at: "http://sccn.ucsd.edu/eeglab/"

Noachtar, S. et al., 1999. A glossary of terms most

commonly used by clinical electroencephalographers

and proposal for the report form for the EEG findings.

International Federation of Clinical Neurophysiology.

Chapter 1.5.

Shamsaei, G.R., n.d. Review Of Clinical

Electroencephalography. Jundishapour University of

Medical Sciences.

Sovierzoski, M.A., Argoud, F.I.M. & Azevedo, F.M.d.,

2008. Identifying Eye Blinks in EEG Signal Analysis.

In IEEE., 2008.

Spriggs, W.H., 2010. Essentials of Polysomnography.

Jones and Bartlett Publishers, LLC.

Stern, J.M., 2005. Atlas of EEG Patterns. Lippincott

Williams & Wilkins.

Wang, P., Li, S., Shao, M. & Liang, C., 2016. A Low-Cost

Portable Real-Time EEG Signal Acquisition System

Based on DSP. International Journal of Signal

Processing, Image Processing and Pattern

Recognition, pp.239-46.

APPENDIX

Matlab code that includes filtering used for

displaying spectrogram screenshots.

fs = 250;

nfl = 49;

nfh = 51;

fl = 49;

fh = 0.5;

order = 2;

%Butterworth notch filter

[bn,an]=butter(order,[nfl

nfh]/(fs/2),'stop');

%Butterworth low pass filter

[b,a]=butter(order,lp/(fs/2),'low');

%Butterworth high pass filter

[b,a]=butter(order,fh/(fs/2),'high');

%Spectrogram

spectrogram(eegdata_f,hanning(256),2

55,[1:40],250,'yaxis');

Detection of Electroencephalography Artefacts using Low Fidelity Equipment