Final

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/321294225
Electroencephalogram
Chapter · January 2018

DOI: 10.1007/978-3-319-69362-0_2
CITATIONS READS
26 30,499
6 authors, including:
Alex Casson Meera Dulabh

The University of Manchester Manchester University
143 PUBLICATIONS 2,914 CITATIONS 2 PUBLICATIONS 52 CITATIONS
SEE PROFILE SEE PROFILE
Siddharth Kohli
8 PUBLICATIONS 123 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Gealtine Head Phantom View project
Sleeping Soundly: Effects of Auditory Closed-Loop Stimulation on Sleep and Memory (PhD thesis) View project
All content following this page was uploaded by Alex Casson on 27 January 2018.
The user has requested enhancement of the downloaded file.

Chapter 2
Electroencephalogram
Alexander J. Casson, Mohammed Abdulaal, Meera Dulabh,

Siddharth Kohli, Sammy Krachunov, and Eleanor Trimble
2.1 Introduction
The electroencephalogram (EEG) is a widely used non-invasive method for mon-

itoring the brain. It is based upon placing metal electrodes on the scalp which
measure the small electrical potentials that arise outside of the head due to neuronal
action within the brain. Its key benefits compared to other brain imaging techniques
are that it has a very high time resolution – able to track events within the brain with
millisecond accuracy – and that it is in principle portable allowing real-world
neuroimaging to be performed outside of clinical and lab environments. As a result
it is a very widely used sensing modality for a range of health and wellbeing
applications ranging from epilepsy diagnosis to emotional monitoring.
This chapter will overview the EEG: its set up, analysis and use. Section 2.2
discusses the principle of the EEG and how it is recorded and analysed, giving a
practical getting started guide. Section 2.3 gives the technical details behind this set
up, with significant attention given to overviewing the state of the art in the type of
recorder and electrodes used. Section 2.4 then covers the typical applications where
EEG is used today and discusses some of those which may become more common
in the future as EEG systems become even smaller, easier to use and more robust to
A.J. Casson (*) • M. Abdulaal • S. Kohli

School of Electrical and Electronic Engineering, The University of Manchester, Manchester,
UK
e-mail: [email protected]
M. Dulabh • E. Trimble
School of Materials, The University of Manchester, Manchester, UK
S. Krachunov
EPSRC Centre for Doctoral Training in Sensor Technologies and Applications, The University
of Cambridge, Cambridge, UK
© Springer International Publishing AG 2018 45

T. Tamura, W. Chen (eds.), Seamless Healthcare Monitoring,
https://doi.org/10.1007/978-3-319-69362-0_2
46 A.J. Casson et al.
interference sources. The chapter concludes in Sect. 2.5 with a summary and
potential future prospects.
2.2 Principle
2.2.1 Introduction
To overview the EEG, it is first necessary to consider, at a high level, the origin of
the signal within the brain, the set up of the recording instrumentation and the
typical signals that are obtained. This section introduces these topics to give a
general background to the more detailed considerations of different electrode and
amplifier types that will be considered in Sect. 2.3.
2.2.2 EEG
2.2.2.1 Origin of EEG
The EEG was first carried out in 1929 by the German psychiatrist Hans Berger [1]
and has a long history of development and use. At its core, the EEG is essentially
the same as the well-known electrocardiogram (ECG) used for monitoring the
heart, but applied to the head. In brief, brain activity is characterised by the passing
of electrical impulses along neurones and postsynaptic responses as neurons com-
municate with one another. Electrodes attached to the head detect the cumula-
tive electric fields associated with these impulses, and the potential differences
produced can be amplified and stored giving characteristic representations of brain
activity.
Research into the precise cerebral origin of EEG signals that manifest outside of
the head is still ongoing [2–4]. It is clear that the brain has a large number of
electrical sources present in it: each neuron has intrinsic electrical properties as
action potentials are generated by voltage-gated ion-channels in the cellular mem-
branes, and synapses operate based upon the flow of sodium and potassium ions.
The scalp EEG is a very large-scale sum of this electrical activity from large
populations of neurons and glial cells operating in synchrony and with volume
conduction effects affecting the size of the brain area to be considered. For practical
use, the EEG can be viewed as an emergent property of these populations and
networks: a voltage waveform with its own characteristic shapes and properties
appears on the scalp due to the neuronal action within the brain, and it is not
necessary to consider the detailed cellular origins to make meaningful use of these
EEG signals.
Nevertheless, substantial work has been carried out towards solving the inverse
problem, that is, identifying electrical sources within the brain based upon the
measured scalp signal [3]. As there are many more electrical sources than
2 Electroencephalogram 47
measurements, this problem is ill-posed and has no unique solution unless addi-
tional constraints are applied. Commonly this is done by assuming a model
whereby EEG activity is due to a finite number of electrical sources known as
dipoles, and the inverse problem seeks to find the contribution of each dipole to the
current EEG trace. Here each dipole is an electrical/mathematical construct to map
the scalp EEG on to, rather than a biological construct giving precise insights into
the underlying cellular level origin. When processing EEG signals (Sect. 2.2.4),
many methods are based upon analysing the time-domain EEG collected on the
scalp directly, while others are based upon estimating the dipoles and making use of
these as the signal processing basis. Both are valid and common EEG analysis
approaches.
2.2.2.2 Typical Signals
The EEG signal that arises on the scalp is measured as a voltage in the time domain,
with a wide number of potential signal morphologies present. Figure 2.1 shows
some example waveforms, although a small number of examples could never
capture the wide range of signals that are seen in practice. An atlas, giving many
more examples, is available in [5]. In general, the EEG is not a nice looking signal.
To the untrained eye, it often looks only like noise, and it takes significant
experience for a human to be able to interpret anything beyond the coarse features
that are present.
These features can be classified in multiple ways, with different methods being
common depending on the field of application. Very common is to divide the EEG
into free-running, evoked and hybrid components.
Free-running EEG is the brain activity that is present due to the normal operation
of the brain. It is there, all of the time, as the brain is operating. This EEG is
characterised by diving it into frequency bands, each given the name of a Greek
letter:
• Delta: Activity at less than 4 Hz
• Theta: Activity between 4 and 8 Hz
• Alpha: Activity between 8 and 13 Hz
• Beta: Activity between 13 and 30 Hz
• Gamma: Activity over 30 Hz
An example of a single EEG trace broken down into these frequency bands is
shown in Fig. 2.2, which illustrates how the different bands evolve over time.
An increase or decrease in the power present in a particular band at any point in
time is then an indicator of the user’s state. For example, when a user is restful and
closes their eyes, a dominant alpha rhythm emerges at the back of the head over the
occipital cortex. This is easily identified and is a common test used to check that the
EEG is set up correctly. Alternatively, the process of falling asleep is associated
with alpha activity being replaced by slower theta activity [6]. Many applications of
EEG are based, at least in part, upon the tracking of these frequency bands over
Fig. 2.1 Examples of EEG signals. (a) Free-running background EEG. (b) An interictal spike, a
feature suggestive of a person with epilepsy
time. Note that it is near universal to divide the EEG spectrum into these bands, but
some EEG disciplines use slightly different frequency ranges [7] as the split points.
Care should be taken to adhere to the standard practice in the precise
application area.
Within free-running EEG there are then a number of features that occur due to
different brain states which will be of interest for different applications. For
example, Fig. 2.1b shows an interictal spike, which is a characteristic feature of
epilepsy and can be used to help locate the epileptic foci within the brain. Focusing
on epilepsy, a number of similar features can be identified:
• Spike: The spike is a classic example of epileptic activity, and its presence is
useful for determining the focus of the epilepsy [8]. Spikes are defined as
Fig. 2.2 Example of EEG channel C4 broken down into frequency bands (using a 211 point fast
Fourier transform) and the tracking of the energy in each of these bands over time
“transient[s], clearly distinguished from background activity . . . [with] a

duration from 20 to under 70 ms” [9].
• Sharp wave: Sharp waves are similar to spikes but with a duration between
70 and 200 ms [9]. Generally, however, they are of less interest when
interpreting EEGs as they provide less localising information [8].
• Spike-and-wave (spike-and-slow-wave): Here a spike is followed by a wave
which has a duration of more than 125 ms [9].
These features, as is common for the EEG in general, can be highly variable. For
example, based on these definitions [10] presented a detailed characterisation of
600 spikes in 120 EEG records from 100 people with epilepsy. Spike durations
varied between 9 and 200 ms with a mean of 45 ms, and 98.3% of spikes had a peak-
to-peak amplitude that was 30% or more above the average background amplitude
from the preceding 500 ms. Further, different spike morphologies and durations
were often seen in the same patient. The main component was negative in 88% of
spikes, and no purely positive spikes were found.
Similar features can be identified for other medical conditions and brain states,
most noticeably sleep analyses which are based upon identifying:
• Slow oscillations: A clear sub 1 Hz oscillation with negative going state,

followed by an upwards going state that lasts for several hundred
milliseconds [11].
• Sleep spindles: “A burst at 11–15 Hz but mostly at 12–14 Hz generally diffuse
but of high voltage over the central regions” [9]. Generally split into fast and
slow spindles depending on the peak frequency.
• K-complexes: “A high voltage negative slow wave followed by a smaller positive
slow wave frequency associated with a sleep spindle” [9]. Noted to be variable in
appearance and not due to external stimuli.
In contrast to free-running EEG, evoked EEG arises due to stimuli being
presented to the user. For example, if a user concentrates on a flashing light at a
particular frequency, that stimulus produces a steady-state visual-evoked potential
(SSVEP) [12]. That is, an oscillation at the same frequency as the light source arises
in the EEG at the back of the head. Similar steady-state responses can be found due
to audio stimuli, in which case they are termed auditory steady-state responses
(ASSR) [13]. These evoked responses form the fundamental basis of many brain–
computer interfaces (BCIs). For example, if a screen has multiple light sources,
each at a different frequency, it is possible to tell which source the user is focusing
on as this will be the frequency of the resulting SSVEP. This information can then
be used to control a cursor moving around the screen or to make a decision.
There are in fact a very wide number of evoked responses that are possible due to
different forms of stimuli and experimental set ups. In addition to steady-state
responses, also common are event-related potentials (ERPs) which arise due to the
presentation of individual stimuli, with a gap present before a subsequent stimuli
presentation. Below are some common ERPs:
• P100: elicited by using checkerboard (alternating black and white) stimulation
• N100: produced by the presence of an unexpected stimulus, particularly audi-
tory, when no other task is being performed
• N170: elicited when a face is present in visual stimuli
• P300: produced by an oddball stimulation, that is, when looked-for uncommon
stimuli are observed in a train of other stimuli
• N400: produced in response to the recognition of a face
All of these are named P, for a positive going deflection, or N, for a negative
going deflection, together with a number which reflects approximately how long
after the presentation of the stimuli the response is evoked (in milliseconds).
Not all of the possible ERPs will be present in every experiment, and the range of
choices available gives a great deal of freedom for investigating different uses of
EEG as a tool for understanding the brain (the precise shape/timing of ERPs is
modulated by the current brain state) and for making BCIs. Figure 2.3 shows an
example of evoked responses produced during a face detection task where the user
was shown a series of pictures for 1 s, followed by 1 s pauses, with the picture
randomly selected to be a recognisable face or static noise. A number of evoked
responses can be seen. Here P100, N170 and P300 responses are present.
Fig. 2.3 Averaged evoked responses at electrode position P08 in a face detection task. A number
of evoked responses are seen compared to a control case where no faces are included in the visual
presentations. The responses are named after their approximate timings, although these can be
highly variable
For making use of such signals, it is important to note that while steady-state
responses are large in amplitude (particularly in the frequency domain), other
evoked responses are often very small. Indeed they can be below the noise floor
of the EEG electrode and amplifier and so are not directly observable in the time
domain. To extract the responses, time-locked averaging is used: in an experiment
the time of each stimuli presentation is known, and this is used to cut the signal into
multiple trials which can be aligned at the stimulation time and averaged together.
As noise present in the system is uncorrelated between trials, it will average to zero,
while the evoked response is present in all trials and will be maintained. The more
trials and averages that are taken, the better the response extraction, at the cost of
speed. For example, if each stimulus/trial must be separated by 1 s to allow all of the
ERPs that it produces to have finished, only 1 in 3 trials randomly includes the
wanted target stimuli, and 10 averages are required to robustly extract the presence
of an evoked response; it takes 30 s to extract one piece of information. Lots of
research effort has been invested in reliably detecting responses from a single trial
and similar methods to speed up the paradigm. In addition, evoked responses are
highly variable. The typical amplitudes and precise timings of evoked responses are
different between different people and vary within the same person, for example,
with age and a number of other factors [14]. There is also evidence that not all
people reliably generate detectable evoked responses, either due to their individual
variability or an incorrect use of the equipment set up [15].
The last class of data, hybrid EEG, is between the above two cases. No direct
stimuli are presented to the user, but they are asked to think of something or
imagine performing an action. This can then result in known signal morphologies
arising in the EEG. The best known such signal is associated with motor imagery:
when a user is asked to imagine performing a hand movement, EEG activity at 8–12
and 18–26 Hz decreases over the motor cortex (around electrode positions C3 and
C4), and there is an event-related desynchronization [16]. This is illustrated in
Fig. 2.4 which shows the event-related desynchronization associated with imagined
Fig. 2.4 Imagined left and right hand movements show event-related desynchronizations (defined
in [105]) on the side of the head opposite to the imagined movement. This can be used as the basis
of a BCI where a user is asked to imagine different movements in order to invoke different
commands from a computer (Image adapted from [105])
left and right hand movements on the different sides of the head. This and similar
hybrid tasks form the basis of many BCIs.
2.2.3 Measurement
2.2.3.1 Basic Set Up
Typical signals detected by scalp-mounted electrodes are in the range 1–150 μV

over a 0.1–60 Hz bandwidth, although some higher frequency content may also be
present [8, 17]. The signals vary both temporally and spatially, and so multiple
electrodes are typically used. Electrode positions are determined using the 10–20
standard, so named as distances between electrodes are measured as being 10% or
20% of the skull dimensions, as illustrated in Fig. 2.5. Electrode letters correspond
to their position on the skull, with odd number electrodes being on the left
hemisphere and even ones on the right. A 10% system is described in [18] and
can be used in situations when further electrode positions are required to get more
coverage over the head. Smaller subdivisions are also possible for when a very large
number of electrodes are used.
A pair of electrodes is required in order to obtain a voltage potential difference,
and each pair of electrodes is thus connected to an amplifier. After suitable
amplification and bandwidth limiting, the signals are stored in a suitable location.
In older devices an analogue signal was stored on a magnetic tape or written out by
a pen writer [8]. Modern devices digitise the signals allowing them to be stored,
wirelessly transmitted or analysed in real time as desired for the particular
application.
At this point, it is important to note that a wide number of variants on this core
set up are available, and sometimes also referred to as the EEG. For example, semi-
invasive electrodes placed not on the scalp, but under it, above the bones of the
skull, have been proposed as an approach for very long-term recordings performed
Fig. 2.5 The standard 10–20 electrode system for electrode placement and names
in intensive care units [19]. Similarly, the Electrocorticogram (ECoG) [2] is

sometimes referred to as depth EEG (DEEG) or incorrectly as simply EEG
[9]. The ECoG uses similar recording principles to the EEG but with electrodes
placed invasively inside the head typically directly on the surface of the brain.
Depth recordings, where microneedle electrodes penetrate the brain, are also
possible. The skull tends to act as a low-pass filter, and ECoG and depth recordings
show much higher frequency and higher amplitude content along with different
artefacts and a higher level of source localisation compared to scalp EEG [8]. As a
result the instrumentation and signal processing requirements for the two situations
are very different, and techniques cannot necessarily be used in both domains. Due
to its invasive nature, the ECoG is not suitable for use in most situations, except in
the most severe of cases. In this chapter we focus on technologies for non-invasive
scalp-based EEG recording.
2.2.3.2 Practical Set Up
Although a wide range of choices are available when performing a modern EEG
recording, the conventional set up that first comes to mind when discussing the EEG
is shown in Fig. 2.6a. This illustrates a user with a head cap on which has holes to
hold a number electrodes next to the scalp. Each electrode has a long wire which
allows it to be connected to recording instrumentation. On each electrode a con-
ductive gel is placed in order to ensure that a good contact is made between the
metal of the electrode and the scalp, and a close-up view of this is shown in
Fig. 2.6b. In conventional EEG this gel is critical to getting a good electrical
contact with the head, and it can act as a mechanical buffer to ensure that the
connection is maintained even during and after head movements.
Figure 2.7 then illustrates a conceptual set up of EEG recording instrumentation.
Electrodes placed on the scalp detect a small electric signal which is amplified and
stored. A differential architecture is used to remove common-mode interference
signals which may couple to the (potentially) long wires coming from the head, and
an explicit ground or driven right leg electrode (as with the ECG, see [20]) is used to
further suppress common-mode interference. A recording from a pair of electrodes
forms one EEG channel, and different montages are possible depending on which
electrodes are used. For example, in a referential montage, the second input to the
differential amplifier is the same for all of the channels, whereas in a bipolar
montage, electrodes are connected in chains (so channels are formed as electrode
1–electrode 2, electrode 2–electrode 3 and so on).
Modern set ups favour referential recordings as this allows any desired other
montage to be derived offline by software [21]. For clinical-grade recordings,
equipment recommendations from the International Federation of Clinical Neuro-
physiology for digital EEG recording are given in [21] and summarised in
Table 2.1.
Fig. 2.6 (a) A conventional EEG set up with metal electrodes on the scalp held in place by a cap.
Long wires connect them to recording instrumentation. (b) Close-up of an electrode making
contact with skin via a conductive gel
Fig. 2.7 A simplified overview of an EEG set up: recording electrodes are attached to a
differential amplifier which is connected to a storage medium. In traditional systems this medium
was a pen writer. In modern systems it is an analogue-to-digital converter and on-board memory or
wireless transmitter
Table 2.1 Summarised standards for clinical-grade digital EEG recording from [21]
At least 24 channels, preferably 32 Interchannel crosstalk less than 1%
(40 dB down or better)
At least 12 bit analogue-to-digital conversion with a 70 Hz, 40 dB per decade, anti-aliasing
minimum resolution of 0.5 μV filter required
Minimum sampling rate 200 Hz, preferably higher Recording noise not more than 1.5 μVpp
or 0.5 μVrms
High pass filtering at 0.16 Hz or less Common-mode rejection ratio at ampli-
fier input at least 110 dB
50/60 Hz notch filter (no specified Q) available but Referential montage to allow later
not routinely used re-montaging
Electrode impedances below 5 kΩ Amplifier impedances over 100 MΩ
2.2.4 Signal Processing and Data Analysis
2.2.4.1 Overview
Analyses of EEG for medical applications are very much based upon a human
scrolling through the collected data to identify and differentiate points that are of
interest. However, the EEG is a very small signal which can be hard to collect, is
often subject to substantial interference (particularly artefacts due to motion) and
can be difficult to interpret reliably even by a human. For example, in EEG-based
epilepsy diagnosis, when experts are asked to mark candidate epileptic events in the
same EEG recording, they will often mark noticeably different sections, with
agreement ranging from 0% to 90% of the time [22]. As a result there has been
very substantial interest in applying signal processing to automate the many
different analyses of EEG data that are possible. It is an ongoing task with much
research activity in this area today. The aims and benefits are generally:
• To speed up analysis compared to needing a human to look at all of the data
• To increase accuracy compared to human interpreters who may overlook fea-
tures of interest
Fig. 2.8 Basic EEG signal

processing begins by
selecting the data section of
interest and cleaning the
EEG to remove interference
and artefacts, followed by
feature extraction,
normalisation and
classification to make a
decision based upon the
data
• To disambiguate analysis, giving a single justifiable interpretation of an EEG

trace and removing scope for debate
• To enable new applications, such as BCIs which do not require a human to
interpret data, and would be infeasible to have responsive systems, or ones based
on difficult to calculate measures, if human analyses were required
• To give novice users access to EEG technology without having to spend several
years becoming familiar with the wide range of signal morphologies that are
present
The short summary of EEG signal processing is thus that: if you name a signal
processing approach, it has almost certainly been applied to the analysis of EEG
signals at some point. As a result it is very difficult to concisely summarise the wide
state of the art available in this area. Nevertheless, although many variants are
possible for different application set ups, Fig. 2.8 gives a brief overview of the
standard signal processing flow that is applied. Below we give a general overview
of these steps, without going into large numbers of variations or into mathematical
details which are very well summarised in [23] for the interested reader.
2.2.4.2 Motion Artefact Removal
The first step is in ensuring that high-quality EEG data, as far as possible, is the
input to the signal processing chain. Figure 2.9 shows a worse case example set of
traces in the presence of bad electrode connections and motion artefacts. Essentially
no information of physiological origin is present in the signal, and so it is not
meaningful to process this directly. In the worst cases, such sections of data are
simply discarded from the analysis. A key aim of longitudinal EEG analysis is to
collect sufficient data to allow the required insights into the brain to be obtained,
even if large sections of data must be removed due to such artefacts. Motion
artefacts can be minimised at the data collection stage by ensuring that the elec-
trodes are correctly and well connected to the head and that any wires are kept short
in length and not allowed to move freely.
Fig. 2.9 Examples of EEG signal artefacts caused by bad electrode contacts which mean large
amounts of mains (50/60 Hz) and motion interference are collected. Note the change in scale
compared to Fig. 2.1
For other sections of data, it is possible to remove motion interference algorith-

mically. Most approaches for removing such interference, and recovering clean
EEG underneath artefacts, are based upon signal decomposition techniques such as
independent component analysis (ICA) and principal component analysis (PCA)
[23]. ICA is particularly popular and implemented in the widely used EEGlab
toolbox for Matlab [24] which allows artefact removal to be performed without
having to be involved with the full mathematical details. ICA assumes that the EEG
signals are made up of a number of non-Gaussian signal components which are also
statistically independent of one another. ICA finds these signal components, and it
is then the case that artefacts, which are statistically very dissimilar to true EEG
components, dominate only one or two of the components. The contaminated
components can thus be discarded, and analysis continued with the remaining
components or a time domain cleaned EEG trace reconstructed by performing an
inverse procedure with the reduced number of components.
Such ICA-based cleaning can be very successful, and a number of groups have
now demonstrated the recording of evoked potentials, below the free-running EEG
noise floor, while subjects are walking on a treadmill [25, 26]. The walking speeds
used in such experiments tend to be slower than real-world walking but demonstrate
substantial potential for use in rehabilitation applications. EEG during cycling has
also been demonstrated [27, 28].
The limitation of ICA- and PCA-based approaches is that their mathematical
conditioning is dependent on the number of EEG channels present. Generally,
around 20 channels are regarded as the minimum required in order to get good
decomposition performance. (Note that there are only 18 channels in Fig. 2.9.)
Many computational neuroscience studies use 64 or more channels as a standard.
For lower channel count measurements, of interest in wearable and quick to set up
situations, the use of empirical mode decomposition has been proposed [29] to
artificially increase the number of inputs that can be passed to ICA. However such
efforts are still very much at the research stage.
2.2.4.3 Data-Driven Machine Learning
Given the EEG input, the remaining steps in Fig. 2.8 are the same as those used in
many data-driven machine learning approaches. The end aim is to use a classifier,
whether a support vector machine, artificial neural network, k-means-clustering or
other method (potentially a simple threshold) to make an end decision about the
current window of data being analysed. This might be to indicate the presence of an
evoked response or to classify a section of data as being associated with a high
workload or a low workload or reporting the current sleep stage. To do this a
number of features are calculated from the data. That is, signal processing is applied
to emphasise the points of interest in the signal to increase the discrimination
between these points and other uninteresting data sections. The list of potential
features and feature extraction methods is essentially limited only by the algorithm
designer’s imagination and will be very particular to the precise application being
pursued. A list of 65 such features is given in [30], and it is also recognised that
non-linear mathematics can provide many methods for discriminating points of
interest from background EEG and residual artefacts [31].
Within this wide array of choices, for completeness it is important to briefly
mention here the three common groupings. Firstly, time–frequency features remain
a very common basis. This allows the extraction of EEG frequency bands (Sect.
2.2.2.2), which are directly used in many applications. The standard fast Fourier
transform (discrete Fourier transform) is commonly applied for this, although it is
recognised that the EEG is nonstationary: the frequencies present in the signal
evolve other time rather than always being present as is assumed with the Fourier
transform. Approaches such as the wavelet transform [32] which overcome this
have thus been explored extensively. Secondly, ICA can be used to calculate the
EEG dipoles used as source models for the EEG voltages [33] (Sect. 2.2.2.2) and
this domain used as the signal processing basis rather than the time or time–
frequency domain. Thirdly, methods based upon common spatial patterns (CSP)
are becoming very common for motor imagery brain–computer interface applica-
tions. The common spatial pattern approach weights different electrodes such that
the variances of the resulting time series are optimal for discriminating between
patterns, with each electrode contributing depending on its importance to the end
classification task [16, 34]. It is one of the key methodologies behind recent
advances in robust motor imagery-based control, particularly of drones [35, 36].
2.2.4.4 Performance Assessment
Finally, attention must be given to the performance assessment methodologies for

EEG signal processing. Applying signal processing to the EEG is straight forwards.
Proving that the signal processing operates as expected and that the reported level
of performance is accurate and repeatable is much more challenging. This is most
keenly seen in epilepsy feature detection applications, which due to their substantial
personal and economic burden were some of the first clinical EEG interpretation
areas to have signal processing applied, going back to the 1970s [37]. However,
despite over 40 years of substantial work, and thousands of academic papers
investigating all manner of different signal processing bases, many with perfor-
mances in excess of 90%, it is widely recognised that few clinical groups routinely
use such algorithms as part of their care procedures. Similar statements can be made
for automatic sleep staging and other clinical applications of EEG signal
processing.
Robustly, repeatably and believably reporting the performance of different
algorithms is extremely challenging due to the wide number of factors that affect
EEG traces. Age significantly affects the EEG amplitude [38]; EEG changes with
tiredness and with alcohol consumption and with different medications, amongst
other factors [39]. As a result the EEG is autocorrelated with itself in time, and
signal processing features can have very different statistical distributions depending
on which recording session they are collected from [40]. Controlling for all possible
factors that affect the EEG is not possible in small-scale developmental studies, and
this can lead to an overestimation of the performance when the developed signal
processing is applied to new unseen data sets. Robust methodologies are application
specific, and best practices have been reported elsewhere [39]. Here, it is sufficient
to note that equal attention must be given to both the signal processing development
and to the performance verification strategy used. Works which overlook the
importance of performance assessment methodology are unlikely to produce results
which are generalizable to EEG data beyond that used in the current specific study.
2.2.5 Summary
This section has given a brief introduction to EEG technology, the typical wave-
forms that are encountered and how they are processed. A wide number of choices
are available. The next section investigates some of these choices in more detail,
particularly with regard to the advances in wearability of the EEG sensing
equipment.
2.3 Modality of Measurement
2.3.1 Introduction
Section 2.2 demonstrated that EEG hardware has a long history – nearly 100 years
of technological development. The result is that a wide number of choices are
available for EEG experiments today in terms of both the electrodes, which are
critical for signal quality and user comfort, and the recording hardware. In general
there is no one perfect EEG system that gives the optimal set up for all situations
and applications. Rather, there is always a trade-off required between the number of
channels, wearability, ease of set up, data quality and similar. This section intro-
duces the state of the art in seamless EEG monitoring informing such trade-off
decisions, focusing particularly on the EEG electrodes, how they attach to the scalp
and the EEG instrumentation.
2.3.2 Electrodes
The choice of EEG electrode is critical for obtaining good-quality EEG signals. The
EEG is very low amplitude, easily corrupted by noise, and without good-quality
electrodes and set up to connect them well to the head, it is very difficult to obtain
high-quality data. Today, there are three main types of electrode available: passive
wet, active and dry.
Figure 2.10 shows a classical, passive wet, EEG electrode in disposable
(Fig. 2.10a) and reusable (Fig. 2.10b) forms. Both are very commonly used, with
the choice determined by whether it is economically worthwhile to spend time
cleaning and sterilising the electrodes after use. Figure 2.10c gives the equivalent
electrical circuit. In many ways these electrodes are unchanged from the early days
of EEG recording: they are a piece of metal attached to a long wire. There is a hole
in the centre of the electrode (or a cup) to allow a conductive gel to be added which
ensures a conductive path is made between the electrode metal and the scalp.
Passive electrodes such as these are very commonly used today and remain the
gold standard in clinical use and in many research labs.
Many conductive materials can be used for the electrodes in order to fundamen-
tally carry out a recording [41], although these all have different properties in terms
of contact noise, long-term drift and similar. The passive electrode is a transducer
which converts ionic currents coming from the human body into electron currents
that can be measured by conventional electronics. In Fig. 2.10c, Vhc represents the
half-cell potential of the electrode, which is a function of the material used and
which occurs in the presence of charge carriers (ions) concentrated at the electrode–
electrolyte interface and which introduces noise and a long-term drift in the
recorded potentials [42]. In Fig. 2.10c, Ri and Ci are parameters which determine
Fig. 2.10 Silver/silver chloride (Ag/AgCl) passive wet EEG electrodes. (a) Disposable. (b)
Reusable. (c) Equivalent electrical circuit
the impedance of the electrode–electrolyte interface, and Re is the material resis-

tance of the electrode.
Although other materials are used, silver/silver chloride (Ag/AgCl) electrodes
are the most common due to their excellent performance in practice. They have a
very low half-cell potential resulting in better drift and noise performance at DC
and low frequency measurements and are nonpolarisable. Polarisation is an
undesired effect which occurs when there is a build-up of charge carriers at the
electrode–electrolyte interface which acts as an insulating barrier such that current
is no longer able to pass and reduces the long-term stability of the electrode
interface. Typical electrodes are about 1 cm in diameter, with the electrode contact
noise being inversely proportional to the electrode diameter [43].
Key to good operation of passive electrodes is to ensure that the contact
impedance between the electrode and the scalp is low, always below 10 kΩ and
ideally below 5 kΩ. Electrode set up is always done while monitoring the connec-
tion quality via an impedance metre, and different levels of hair parting, skin
cleaning, top layer skin abrasion, adding gel and compacting gel to make good
contact can be carried out to bring the impedance to an acceptable level. (See also
Sect. 2.3.3.) Typically impedance measurements would be done at the start and end
of an EEG recording to help confirm the electrodes have been attached throughout
the data collection, noting that impedances typically fall after initial connection
Fig. 2.11 Active EEG

electrodes where a buffer
amplifier is placed
immediately on top of the
Ag/AgCl sensing contact
(as the half-cell chemical reaction stabilises) and then increase over time as the gel
dries out and the electrode contact quality reduces.
Figure 2.11 shows the second electrode type: active electrodes. Here part of the
instrumentation electronics, principally a buffer amplifier, is placed on top of the
electrode itself. The obvious drawback is that it increases the electrode size and
weight. However, it has the advantage of reducing mains interference and artefacts
due to the movement of the recording wires. With the passive electrodes from
Fig. 2.10, the entire cable is connected to the high input impedance amplifier input.
These cables are long and can pick up a large amount of interference from the
electrical mains. As the impedance is high, small amounts of picked-up current,
essentially due to a loop area and the Biot Savart law, result in a large interference
voltage as the cables move through the electrical field due to the mains. In contrast,
if a buffer amplifier is included, the cable is connected to the low-output impedance
buffer and the same amount of induced current produces a much smaller interfering
voltage. The high impedance node between the amplifier and the electrode is made
to be much shorter (just the height of the electrode) minimising the path for
interference to be introduced.
A number of variants on this basic arrangement, varying the gain provided at the
electrode in order to get the best trade-off between noise and power consumption,
have been proposed [44]. These electrodes generally still require a gel to be added
but much less than in the classical passive case as high-quality recordings can be
obtained with such a set up even with impedances up to 40 kΩ present [45].
Figure 2.12 shows a selection of examples from the final electrode type: dry
electrodes. These attempt to overcome the need to have a conductive gel present.
Although very important for getting the best signal quality, this gel takes a long
time to apply, leaves a mess, dries out over time and is highly unpopular with both
users and researchers. There have been many attempts to make electrodes that do
not require it, with academic papers dating back to at least 1994 [47]. Taheri et al.
Fig. 2.12 A selection of dry EEG electrodes which do not require a conductive gel to be present.
(a) Cognionics (http://www.cognionics.com/). (b) g.tec (http://www.gtec.at/). (c) Wearable sens-
ing (http://www.wearablesensing.com/). (d) 3D printed [46]
[47] presented a dry active electrode for EEG recordings based on a 3 mm steel disc
coated with nitride on one side and an impedance-converting amplifier on the other.
The key challenge in designing dry EEG electrodes, without the gel bridge
present, is that it is much more difficult to keep the electrodes in place and next
to the scalp for long-term recordings. Most dry electrodes focus on having fingers as
shown in Fig. 2.12 in order to better penetrate the hair and are attached to springs to
help keep them in place. In addition, the electrode contact impedance is much
higher, and this necessitates careful design of the EEG amplifier and PCB in order
to avoid introducing contact noise and motion artefacts. Research into these factors
is still ongoing [48]. Nevertheless, today in 2017, dry electrodes are available
commercially from a number of suppliers and can also be 3D printed to allow
easy access and personalisation for each individual [46]. Current designs can
certainly be used to record EEG signals without requiring a gel but are unlikely
to be the electrode of choice for when the highest-quality recordings are required.
Part of the challenge is that there is little standardisation on how to characterise dry
electrode properties [49], and direct comparisons between designs presented in
different papers are difficult to do accurately. A recent review comparing different

types of dry EEG electrode is given in [50].
2.3.3 Electrode Attachments to the Scalp
Most overviews of EEG electrodes focus on the choices given in Sect. 2.3.2, that is,
the choice of gel [41], the electrode material, whether a buffer amplifier is included
and the physical arrangement of fingers in dry electrodes. Equally important
however is to give attention to the scalp preparation and how the electrode is
physically held in place. Many poorer electrodes can get better performance simply
by pressing them against the body more forcefully, at the obvious cost of comfort. It
is not a large area of investigation, and it is more that each lab has its own internal
style which works for them. Nevertheless, several choices exist.
For long-term clinical-grade recordings, typically up to 24 h in a single session,
electrodes can be glued in place using collodion. This gives a very high-quality
electrode contact, which is relatively inconspicuous for portable use, but one which
is very difficult and uncomfortable to remove. A tape measure and head crayon are
used to work out the 10% and 20% positions on the head (see Fig. 2.5), a mark is
placed on the scalp and the electrode placed over this before applying the gel
and glue.
More commonly in research set ups, as shown in Fig. 2.6, an electrode cap is
used. Such caps are quick and easy to use and have predefined holes to place each
electrode accurately. Many different sizes and cuts of cap are available to account
for different people/head sizes, head shapes and ethnicities. However, in many
cases the cap will not be fully flush with the head in every place, instead having an
area that bags away from the scalp due to excess material or hair being present.
Electrodes in these areas are not held against the scalp, and it can be very difficult to
get the electrodes to connect in these regions. The caps also tend to be limited to a
few hours of recording time as the tight chin strap becomes uncomfortable, and the
caps are prone to slight movements that disconnect the electrodes.
An alternative to these approaches is to use a conductive gel which is also
adhesive, such as Grass Technologies EC2 [51]. Typically a cap is used to mark
the electrode locations on the head, and then removed, and the EC2 doubles up as
the conductive recording gel and adhesive. This takes more skill to administer than
a plain cap, but can give very good results, and in our lab is the preferred choice
unless there is a particular reason to use a cap.
As will be seen in Sect. 2.3.4, EEG units are also becoming increasingly
available as headsets (see Fig. 2.13) as predefined sizes with plastic arms extending
to different parts of the head from a central electronics housing. These headsets
have the advantage of being more aesthetically pleasing, not covering the entire
head, and being more comfortable than caps as no chin strap is required. Nearly all
modern wearable units take this approach. The downside is that, while the arms are
designed to be flexible in order to accommodate different head sizes and shapes, on
Fig. 2.13 An overview of EEG modalities, with reducing size and increased portability (Ear-EEG
picture originally from [52])
any one person it can be very difficult to get all of the arms to actually contact all of
the parts of the head at the same time. In some cases it can be extremely difficult to
get all of the electrodes to connect with the scalp.
For all of the above, for good-quality recordings scalp preparation is key. It is
possible to put a unit on and then adjust and adjust the electrodes to eventually get a
good contact. However, this takes a significant amount of time, experimenter skill
and cooperation from the user. It is much better to prepare the scalp first. A simple
cleaning of the scalp under each electrode location with a q-tip and mild alcoholic
rub will drastically improve the connection quality, reduce impedance and reduce
the set up time. The same can be said for parting the hair in advance. For most hair
types, this is easily achieved using a q-tip, and the aim should be for electrodes to
make contact with the scalp directly, rather than for them to sit on top of the hair
with a large layer of gel going through to the scalp.
2.3.4 Instrumentation
2.3.4.1 Overview
Figure 2.13 shows the evolution of EEG instrumentation hardware from large
research lab/inpatient units to highly miniaturised wearable units and research
based units that go beyond current wearables in order to increase the invisibleness
of EEG devices. Each of the steps in this evolution is discussed in detail below.
2.3.4.2 Research Lab/Inpatient
Inpatient units are large, and non-portable, but very high quality and flexible in their
use. They can be used for anything from a standard short EEG test for clinical use
which lasts between 20 and 30 min [53], up to a few hours for experiments in a
research laboratory, to overnight in the case of sleep study recordings and poten-
tially up to several days for long-term inpatient monitoring [54]. However, this
latter option requires considerable resources in a clinical setting and is not univer-
sally available [54].
These durations are dictated by the needs of the study and the willingness of
participants to take part over the required duration. For example, in a 20–30 min
EEG test for epilepsy, assessment indicative activity is shown by approximately
50% of people with epilepsy [55]. The yield can be increased to around 80%
through the use of multiple tests. In sleep testing, daytime sleep tests are used,
but generally diagnosis relies on a complete night of recording allowing the number
and duration of sleep cycles and awakenings to be investigated.
The electrode set up is typically that given in Fig. 2.10, with long wires
connecting the electrodes to an amplifier box, which is generally physically large.
Modern systems commonly include video monitoring which allows the subject’s
behaviour and any clinical manifestations to be compared to the underlying EEG,
very useful for interpreting the collected data. The key disadvantage of this
approach is that the subject is essentially tethered to the amplifier and can only
move within a very small distance of it. While the electrodes might be temporarily
disconnected from the amplifier to allow the user to use the toilet and similar, the
room/location of the recordings is essentially fixed.
2.3.4.3 Ambulatory
Since the 1980s the miniaturisation of electronics has allowed all of the EEG
instrumentation to be put into a portable unit, allowing subjects to move about
while being recorded and so not being required to stay immediately next to an
amplifier box. This method is known as ambulatory EEG (AEEG). A typical
modern unit is shown in Fig. 2.13, and this might be worn on the back of the
head or on a belt lower down the body. Originally, only a few EEG channels (3–4)
would be recorded on to a cassette tape [56]. Today’s units use solid state memory,
potentially with real-time wireless transmission of the data. They are readily
available with high channel counts, with some compromises to maintain battery
life, but negligible reduction in data quality compared to using an inpatient system.
The main limitation is that they have fewer auxiliary ports for trigger inputs for
synchronisation, simultaneous video and similar simultaneous sensing of other
modalities (e.g. heart rate) compared to inpatient units.
Clinically, the key advantage of AEEG is that it can be used as an outpatient
monitoring arrangement, which is cheaper than inpatient monitoring. It is estimated
that 24 h of ambulatory monitoring is more than 50% cheaper than 24 h of inpatient
monitoring [54]. It also allows the patient to be monitored in their natural environ-
ment, of use for research studies where the subject might be biased by being in a lab
environment that they are not familiar with. Overall in epilepsy it is estimated that
AEEG is clinically useful in 75% of the patients it is used with, and abnormalities
are found in 12–25% of AEEGs for which an inpatient EEG was normal or
non-diagnostic [54]. This is despite the fact that AEEG recordings are likely to
have more artefacts present, for example, due to motion, as the subject is in an
uncontrolled environment. For most research studies which use an AEEG monitor,
the subject would still be asked to be generally stationary. In addition, size, weight,
portability and social acceptability remain as limiting factors.
2.3.4.4 Wearable
Wearable EEG attempts to overcome the above limitations of AEEG and creates
ever smaller and more discrete EEG units that are quick and easy to set up by
non-trained users. There is no concrete line for when an ambulatory unit becomes a
wearable one, [57] puts it as “Wearable EEG is envisioned as the evolution of
ambulatory EEG units from the bulky, limited lifetime devices available today to
small devices present only on the head that can record EEG for days, weeks, or
months at a time.” A number of such units are now available commercially, as
shown in Fig. 2.13 normally based upon using headsets with arms that wrap around
the head, rather than needing a full head cap.
Generally the focus of wearable devices is on enabling real-world neuroimaging
in places and situations where conventional EEG is just not possible. The focus is
not to perform identical monitoring to that used for inpatient studies which need,
and get, the very best data quality. Rather, often there is a requirement for the
equipment to be set up by a non-specialist user or to be used in environments where
lots of motion is present. Thus a focus might be given to low channel counts
allowing quicker set ups and the collection of information at key points around
the head, rather than a full head montage. There is also significant emphasis on ease
of use, and a number of EEG units targeted at consumers are now available, for
example, the Emotiv [58], Muse [59] and Neurosky [60] units, for performing
general purpose EEG. For non-trained users working in specific applications, there
are also specialist units optimised for those applications, such as the Rythm headset
[61] and Kokoon headphones [62] for consumer sleep monitoring. Many questions
have been raised about the data quality from consumer-grade headsets such as these
[63], but there is no doubt they have been very successful at giving easier, and lower
cost, access to EEG monitoring. There is even a widely recognised open source
initiative – OpenBCI [64] – to help make EEG technology widely and freely
available.
2.3.4.5 Beyond Wearables
Inevitably the wearable EEG devices available today still have a number of
limitations [65], and research is ongoing to produce units that are even easier to
put on, give better data quality with robustness to motion artefacts, have better
battery life and more functionality and enable new applications for clinicians and
lay users of EEG technology. There are two key directions in this continuing drive
beyond current wearables.
Firstly, an emerging research direction is in making better use of EEG collected
from non-haired regions on the head. As discussed in Sect. 2.3.3, getting electrodes
attached through the hair is a challenging task. It is the key factor that slows down
putting an EEG unit on and means that it can only be done by a trained person. In
recent years it has been demonstrated that EEG components can be recorded from
the forehead [66], behind the ear in locations covered by standard hearing aids [52]
and from inside the ear canal [67]. Historically, many of these locations have been
considered to be purely artefactual, and, for example, the mastoid behind the ear has
been a common location for ground/reference electrodes as it is electrically very
quiet compared to the rest of the head. Nevertheless, advances in signal processing
have allowed information to be extracted from these locations. For example, [68]
demonstrated ASSR measurements from in-ear electrodes. Locations such as
around the ear are particularly promising as many people are already used to
wearing hearing aids in these places, overcoming some of the social acceptability
issues of placing electrodes in clearly visible locations on the head.
Secondly, the instrumentation electronics are becoming increasing miniaturised.
A particularly promising research direction is that of temporary tattoos, as
pioneered by the Rogers group at Northwestern University [69]. Unlike wearable
devices, conformal temporary tattoo-based devices use very flexible,
non-permanent, substrates similar to those given to children as rub on tattoos.
These have in-built adhesive to connect directly to the skin without requiring a
gel, cap or similar. The result is that the tattoos maintain a high-quality connection
to the body for many days at a time, and because they are highly flexible, they can
follow the contours of the skin and get a much larger contact area compared to a
bulk metal electrode. This gives better quality signals and maintains signal-to-noise
ratios even at very small sensor sizes. As such they intrinsically overcome the
limitations of wearable devices, giving better signal quality, longer-term connec-
tions to the body and a more discrete profile for better social acceptability. The
Fig. 2.14 Example of a temporary tattoo electrode, in this case sized for heart monitoring via the
ECG. In principle if placed on the head, it would record EEG. It shows that high signal-to-noise
ratio (SNR) recordings can be performed over long periods (5 days here), unthinkable with other
electrode techniques (Results originally reported in [71])
Rogers group has demonstrated EEG recordings from the forehead using this
technology [69] and from the ear auricle [70]. In this latter case recordings were
recorded for 2-week periods.
An example of our own work towards such long-term sensing is summarised in
Fig. 2.14 which shows an example of a tattoo-based electrode. These are currently
designed and applied for ECG monitoring of the heart but in principal can be used
interchangeably on the chest for ECG and on the head for EEG. They were
manufactured using an in-house process with three layers of electrically conductive
silver nanoparticle paint, using similar procedures to [72] previously used for
printing antennas [73]. The printing is done on temporary transfer tattoo paper
which is commonly given to children, with excellent biocompatibility and longev-
ity. After the paper backing material is removed, the printed metal electrode is left
behind, stuck to the body by a very thin, approximately 10 μm, layer of plastic,
which provides adhesion under the entire electrode surface, key to maintaining a
good body contact. To date, for heart monitoring as shown in Fig. 2.14, these have
been used for up to 5 days of recordings with high signal-to-noise ratios (SNRs)
maintained throughout. They have yet to be applied for monitoring the EEG, but if
the technology can be successfully mapped from heart monitoring (ECG), these
electrodes offer the potential to make a step change in monitoring durations
compared to current EEG methods.
However, at present, as with the forehead- and ear-EEG approaches, temporary
tattoos are restricted to recording in hair-free parts of the head. While it is now clear
that a lot of information can be extracted from these locations, it is unlikely that
they could be suitable for all EEG applications, as many EEG applications are
reliant on recordings from haired regions. For example, epilepsy diagnosis relies on
recording a full head montage with electrodes in all locations in order to help
localise the epileptic foci. Sleep staging analyses ask for at a minimum C3 and C4
[74], over the centre of the head. Brain–computer interfaces based upon visual
stimuli rely on electrodes placed on the back of the head over the occipital (O) and
parietal (P) electrode locations (see Fig. 2.5), while BCIs based on motor imagery
place electrodes over the motor cortex (around electrode positions C3 and C4). For
most people all of these locations are characterised by large amounts of hair being
present. Devising beyond wearable technologies for haired regions remains an open
research challenge.
2.3.5 Summary
EEG systems have changed radically in recent years, from large non-portable
systems to highly miniaturised ones that can be placed on or in the ear to give
very inconspicuous EEG monitoring, with the aims of being more socially accept-
able, quicker to set up and easier to use by non-specialists. This development
enables a wide number of established and emerging EEG applications, which will
be explored in Sect. 2.4.
2.4 Applications
2.4.1 Introduction
The EEG has a long history, predominately originating with clinical applications in
epilepsy and sleep disorder diagnosis. Recent developments in cheap and portable
EEG units have allowed the creation of consumer neuroscience applications for the
first time where real-time insights into the current state of the brain can be used to
enable a number of novel applications. This final section overviews the main
current applications of EEG technology in and out of the lab.
2.4.2 Medical, Clinical and Neuroscience Research

Applications
The EEG has its roots in medical instrumentation, and this is still the widest
application domain for the technology today. This is based upon its potential for
multi-hour recordings with high time resolutions. Sampling rates are typically
between 200 and 1000 Hz which allows rare transient events such as epileptic
seizures or sleep abnormalities to be captured in a way not possible with other brain
imaging techniques such as MRI which may take several seconds per measurement.
Epilepsy is frequently described as the most common serious neurological
condition [53, 75], although incidence (cases per year) and prevalence (total
number of cases) figures vary considerably between different studies [75] (the
discrepancies being attributed to inclusion criteria and the difficulty of diagnosis).
In developed societies the overall incidence is around 50 cases per 100,000 people
per year, and the prevalence rate is around 10 people per 1000 [75]. In the UK this
corresponds to approximately 96,000 people who require hospital-based treatment,

of which 15,000 will have more than one major seizure a month, and 12,000 who
may require institutional care [75]. In the USA, approximately 10% of the popula-
tion experience a single seizure during their lifetime [54], and approximately two
million people have the disorder with 100,000 new cases each year [76].
The commonly known aspects of epilepsy are undoubtedly the most dramatic
ones: generalised seizures where the sufferer falls to the ground and has rhythmic,
uncontrolled limb movements. This is often thought to be brought on by exposure to
flashing lights (photosensitivity) which may arise from a disco or a standard TV or a
computer game. This symptom is undoubtedly a part of epilepsy, but epilepsy is in
fact a very broad area with many different aetiologies and manifestations. At a
broad level epilepsy is defined as: “A chronic condition characterised by a tendency
to develop recurrent unprovoked seizures” [53]. The most serious consequence of
any disease is death, and epilepsy shows a mortality rate (the ratio of people with
the disorder who die compared to the standard for the age group) of 5.1 in the first
year following diagnosis [53]. This then falls to 3.1 after 5 years [53].
Overall, epilepsy can be very difficult to diagnose correctly, and the EEG is a
key tool in the diagnosis process as seizures manifest with identifiable waveforms
in the EEG, Fig. 2.15. A trained interpreter can assess these and use the information
provided to inform the type of epilepsy that is present, and its origin within the
brain. As noted in Sect. 2.2.4, although lots of effort has been invested in automat-
ing the detection of seizures, the sheer range of seizure morphologies and EEG
variances that are possible mean that such automation remains an unresolved
challenge. This is despite large disagreements between EEG interpreters when
asked to analyse the same EEG section [22]. Importantly, it must always be
remembered that the EEG will only be one part of the diagnostic picture, which
will also include seizure diaries, family histories, blood tests and others [77]. As
[78] recommends, care must be taken to not overweigh the contribution of any one
Fig. 2.15 An example EEG morphology showing epileptic seizure onset as time 2 s
piece of evidence, and if diagnostic doubt remains, it is perhaps best to allow the
passage of time [53]. The EEG is highly useful but not a panacea.
In sleep disorder diagnosis, the gold standard is to use polysomnography which
records the EEG, heart rate, breathing rate, blood oxygenation and eye/leg move-
ments during a full night of sleep. (Wearing all of these equipment does not make
for a very comfortable night of sleep!) For analysis well-defined standards have been
in place for many years [74, 79] for classifying the EEG data into different stages:
• Wake
• Stage one (sleep onset)
• Stage two
• Stage three/four (slow wave/deep sleep)
• REM (rapid eye movement) sleep
• Movement time
The EEG data is split into 30 s epochs, and the interpreter identifies features
according to the scoring manual, and those defined in Sect. 2.2.2.2, to classify each
epoch into one of the above stages. In healthy sleep a person cycles through the
stages 3–4 times in one night, with cycles shortening in duration over the night.
Deviations from this general pattern allow underlying pathologies to be
investigated [80].
Closely linked to this is the automatic detection of sleep onset/drowsiness from
the EEG. Reliable and robust detection and prediction of human drowsiness are of
huge importance for reducing operator error and improving performance. For
example, 20% of road accidents involving serious injury are sleep related [81],
and these can potentially be avoided by detecting drowsiness and lapses of con-
centration and alerting the driver to these. Such drowsiness detection devices could
also be of significant use in other safety critical situations, such as for pilots, power
plant monitoring personnel and military personnel.
With the increasing availability and ease of use of EEG equipment discussed in
Sect. 2.3, in recent years a wide number of other medical applications of EEG have
begun to emerge. Particular examples are in brain–computer interfaces for pros-
thetic control [82] and stroke rehabilitation [83]. There is also increasing evidence
that EEG biomarkers could indicate the development of mild cognitive impairment
and dementia [84]. There are many brain-related disorders to which EEG neuro-
imaging could provide insights, although these have yet to become established
clinical practice, and such wider use of EEG is expected to increase significantly in
the medium-term future.
2.4.3 Brain–Computer Interfaces (BCIs)
The most widely developed non-directly medical applications of EEG technology

are in brain–computer interfaces, which allow the operation of a computer to be
changed or adapted without having to use standard interfacing methods such as
mice, keyboards or voice. Works on such interfaces have been ongoing for over
25 years [85], and a wide number of different paradigms are available [86] making
use of the different free-running, hybrid and evoked EEG features described in Sect.
2.2.2.2. Common methods include P300 spellers where the user is presented with a
flashing grid of letters allowing them to enter words, SSVEP-based buttons where a
number of images (e.g. arrows for directing a cursor) are presented flashing at
different frequencies, and motor imagery where the user imagines a left arm
movement to move a device to the left (and vice versa) [36]. This latter paradigm
has obtained particularly good accuracy in recent years.
There are many applications of such BCI technology, particularly for assisting
subjects who are paralysed and who might not be able to control a computer or a
powered wheelchair using conventional interfaces [87]. The challenge in doing this
is in ensuring that the developed systems are accurate, fit for purpose and actually
codesigned with the targeted end users. It is now widely recognised that many BCIs
are not developed in consultation with users and as such may not be representative
of the needs and typical signals from the target user groups [88]. Instead systems are
commonly designed with users who are young, healthy and easily available on
University campuses. The potential benefits of BCIs for this user group are much
more debatable. While there is undoubtedly a certain novelty in allowing devices,
drones and similar to be controlled using only thought, as discussed in Sect. 2.2.2.2,
many evoked responses are very slow which results in a slow information transfer
rate. Although some high rates have been reported [89] in BCI keyboard applica-
tions, the typing rate is typically only a few characters per minute [90], and this is
after having to take time to set up the EEG equipment, to train the signal processing
classifiers for the current user/session and to deal with user complaints of discom-
fort/fatigue after long sessions [91]. It is difficult to imagine a heathy user with a
range of computer input options available to them going to the effort of making
current BCIs their preferred input choice.
Nevertheless, it is widely recognised that there is substantial potential in BCIs,
and work is ongoing to identify “killer apps” which add meaningful value to the
user by providing information which is not available through other routes. Real-
time workload monitoring and classification are quickly emerging as one potential
area, as it can be based upon free-running EEG allowing frequent update rates
compared to using evoked responses, and it is difficult to estimate using other
sensing methods. At the extreme fatigue end, sleep onset is characterised by a
reduction in alpha activity, which is replaced by theta activity (Sect. 2.2.2.2). There
are numerous papers (e.g. [40, 92, 93].) demonstrating that frequency band changes
can be used to identify less extreme changes in vigilance level. This can then be
used to detect when an operator is in a high or a low workload state to potentially
change the speed at which information is presented. This allows the workflow and
operating environment to be optimised in a real-time and time-varying manner
[94]. It can also be used to enhance human training: the mental load of a new task
can be objectively measured and training times increased or decreased to end the
training only when the task involves a low level of effort [95].
2.4.4 Consumer Neuroscience
Finally, a wide number of out of the lab applications of EEG technology are
currently in development as part of the trend towards personalised and preventative
healthcare. Although to date the most successful wearable monitors have been
“fitbit type” ones for activity monitoring and “Apple Watch type” ones for heart
rate monitoring, these are only a very small subset of all the physiological param-
eters that may be of interest. The wearable EEG units discussed in Sect. 2.3.3 have
enabled and encouraged a wide number of consumer driven applications, and the
market for such neurotechnology is estimated to be worth over $13B by 2019 [96].
Such neurotechnology ranges from drowsiness detectors for car drivers men-
tioned in Sect. 2.4.2, to at home sleep monitors as mentioned in Sect. 2.3.4.3, to
assisting with meditation [97]. In addition, considerable attention has been given to
neuromarketing applications of the EEG as it potentially allows subconscious
decision-making processes to be extracted [98], and this can be done in a portable
way as a user moves around a shop. Many academic papers have shown that it is
possible to extract correlates of emotion from the EEG band powers [99].
Most of these highly novel applications of EEG technology are still at the
research stage. Critical to their wider take up is the believability of their results
and outputs. Being based upon highly portable instrumentation and dry electrodes,
while a subject is nonstationary for significant parts of the experiment, the EEG data
quality is not the same as that obtained from traditional highly controlled inpatient
recording set ups. This has to be compensated for with good testing methodologies
(Sect. 2.2.4.4) or else it is easy to overstate results early on and lose credibility. As a
nascent field, it is inevitable that some false starts will be present, and it is important
that these do not undermine the creation of true “killer apps” which really exploit
the latent impact that real-world neuroimaging would allow.
2.4.5 Summary
The EEG has its core applications in medical diagnoses of epilepsy and sleep
disorders and in fundamental neuroscience experiments. Using new portable hard-
ware, many different brain–computer interface and consumer neuroscience appli-
cations, for example, drowsiness detectors and neuromarketing, are now being
investigated. The key challenge is in devising “killer apps” which are only possible
because of the use of EEG, rather than applications where the EEG could be used,
but does not add any real benefit compared to more easily available and useable
approaches.
2.5 Conclusions and Future Prospects
The EEG is a very widely used technology for neuroimaging. It is unique amongst
sensing methods in that it can monitor the brain portably, over a long period of time,
and with a high time resolution for capturing rare and transient events. As a result it
has seen substantial use in medical diagnoses and increasingly in out-of-the-lab
brain monitoring. In recent years this has been driven by the creation of wearable
EEG devices which are small, discrete and socially acceptable, and work is ongoing
to create even more miniaturised devices which can record EEG signals from the
ear, ear canal and forehead. Combined with advances in temporary tattoo structures
for conformal electronics, there are now demonstrations of EEG recordings lasting
up to 2 weeks at a time, a timeframe unthinkable a few years ago. Nevertheless,
much work remains to demonstrate the ease of use, to quantify the sensing perfor-
mance/noise of new electrode structures and to optimise such new approaches to
EEG instrumentation. There are many emerging consumer neuroscience applica-
tions which will benefit from these technology developments. Drowsiness detec-
tion, emotion tracking and workload monitoring are just the start, and there is
significant scope present for much innovation in the coming years.
Within medical environments the key future prospect is to move beyond simply
monitoring the EEG and to making more use of it to enable real-time actuation and
interventions. As with many current wearable sensors, the EEG measurement is
presently a one-way street for data generation only. Next-generation EEG will
focus on enabling two-way streets that both sense and actuate. These will provide
data-driven feedback that leads to the right treatment at the right time and could
minimise the amount of treatment required and make it timelier and more effective.
Such closed-loop data responsive treatments are beginning to emerge, with adap-
tive deep brain stimulation systems for Parkinson’s disease [100], closed-loop
electrical stimulators for epilepsy [101] and GlaxoSmithKline recently making
substantial investments in closed-loop non-pharmacological electroceuticals
[102]. However, these current approaches are focused on implanted sensors,
which have very different signal properties (larger, higher bandwidth, less noise)
and interference sources (less motion, less 50 Hz pick-up, fewer nearby devices)
compared to the scalp EEG.
Although effective, the costs and practicalities of surgery mean that implanted
devices will inevitably be limited to a subset of the most serious cases, and there are
many opportunities for similar, non-invasive devices to act as an intermediate
between current pharmacological approaches and highly invasive surgery. Design-
ing closed-loop interventions is incredibly challenging and presents regulatory
hurdles but is a key future direction for EEG electrode and instrumentation
research. For the EEG this most likely takes the form of transcranial electrical
stimulation [103], which is based upon injecting currents into the head. It thus
naturally complements the EEG which measures electrical voltages reflecting
temporal changes in the electrical state of neurons and current flows, which are
directly modulated when applying transcranial electrical stimulation. However,
present transcranial stimulation introduces significant artefacts into simultaneous

EEG recordings, preventing neuroimaging at the same time as stimulation
[104]. Work is ongoing to overcome these issues and move towards future
closed-loop EEG systems.
References
1. Berger, H. (1929). Uber das eletrenkephalogram des menschen. Archiv f€ ur Psychiatrie und
Nervenkrankheiten, 87(1), 527–570.
2. Buzsaki, G., Anastassiou, C. A., & Koch, C. (2012). The origin of extracellular fields and
currents—EEG, ECoG, LFP and spikes. Nature Reviews Neuroscience, 13(6), 407–420.
3. Lopes da Silva, F. (2009). EEG: Origin and measurement. In C. Mulert & L. Lemieux (Eds.),
EEG – fMRI (pp. 19–38). Heidelberg: Springer.
4. Jackson, A. F., & Bolger, D. J. (2014). The neurophysiological bases of EEG and EEG
measurement: A review for the rest of us. Psychophysiology, 51(11), 1061–1071.
5. Krauss, G. L., & Fisher, R. S. (2006). The Johns Hopkins atlas of digital EEG: An interactive
training guide. Baltimore: Johns Hopkins University Press.
6. Lal, S. K., & Craig, A. (2002). Driver fatigue: Electroencephalography and psychological
assessment. Psychophysiology, 39(3), 313–321.
7. Curio, G. (2000). Ain’t no rhythm fast enough: EEG bands beyond beta. Journal of Clinical
Neurophysiology, 17(4), 339–340.
8. Binnie, C. D., Rowan, A. J., & Gutter, T. (1982). A manual of electroencephalographic
technology. Cambridge: Cambridge University Press.
9. Noachtar, S., Binnie, C., Ebersole, J., Mauguiere, F., Sakamoto, A., & Westmoreland,
B. (1999). A glossary of terms most commonly used by clinical electroencephalographers
and proposal for the report form for the EEG findings. In G. Deuschl & A. Eisen (Eds.),
Recommendations for the practice of clinical neurophysiology: Guidelines of the interna-
tional federation of clinical physiology, Electroencephalography and clinical neurophysiol-
ogy supplement (Vol. 52, 2nd ed., pp. 21–41). Amsterdam: Elsevier.
10. Celesia, G. G., & Chen, R.-C. (1976). Parameters of spikes in human epilepsy. Diseases of the
Nervous System, 37(5), 277–281.
11. Massimini, M., Huber, R., Ferrarelli, F., Hill, S., & Tononi, G. (2004). The sleep slow
oscillation as a traveling wave. The Journal of Neuroscience, 24(31), 6862–6870.
12. Muller-Putz, G. R., Scherer, R., Brauneis, C., & Pfurtscheller, G. (2005). Steady-state visual
evoked potential (SSVEP)-based communication: Impact of harmonic frequency compo-
nents. Journal of Neural Engineering, 2(4), 123–130.
13. Lins, O. G., & Picton, T. W. (1995). Auditory steady-state responses to multiple simultaneous
stimuli. Electroencephalography and Clinical Neurophysiology, 96(5), 420–432.
14. Truccolo, W. A., Ding, M., Knuth, K. H., Nakamura, R., & Bressler, S. L. (2002). Trial-to-
trial variability of cortical evoked responses: Implications for the analysis of functional
connectivity. Clinical Neurophysiology, 113(2), 206–226.
15. Allison, B., Luth, T., Valbuena, D., Teymourian, A., Volosyak, I., & Graser, A. (2010). BCI
demographics: How many (and what kinds of) people can use an SSVEP BCI? IEEE Trans-
actions on Neural Systems and Rehabilitation Engineering, 18(2), 107–116.
16. Wang, Y., Gao, S., & Gao, X. (2005). Common spatial pattern method for channel selection
in motor imagery based brain-computer interface. IEEE Engineering in Medicine and
Biology Society, 5, 5392–5395.
17. Ebner, A., Sciarretta, G., Epstein, C. M., & Nuwer, M. (1999). EEG instrumentation. In
G. Deuschl & A. Eisen (Eds.), Recommendations for the practice of clinical neurophysiol-
ogy: Guidelines of the international federation of clinical physiology,
Electroencephalography and clinical neurophysiology supplement (Vol. 52, 2nd ed.,

pp. 7–10). Amsterdam: Elsevier.
18. Klem, G. H., Luders, H. O., Jasper, H. H., & Elger, C. (1999). The ten-twenty electrode
system of the international federation. In G. Deuschl & A. Eisen (Eds.), Recommendations for
the practice of clinical neurophysiology: Guidelines of the international federation of clinical
physiology, Electroencephalography and clinical neurophysiology supplement (Vol. 52, 2nd
ed., pp. 3–6). Amsterdam: Elsevier.
19. Martz, G. U., Hucek, C., & Quigg, M. (2009). Sixty day continuous use of subdermal wire
electrodes for EEG monitoring during treatment of status epilepticus. Neurocritical Care, 11
(2), 223–227.
20. Webster, J. G. (1984). Reducing motion artifacts and interference in biopotential recording.
IEEE Transactions on Biomedical Engineering, 31(12), 823–826.
21. Nuwer, M. R., Comi, G., Emerson, R., Fuglsang-Frederiksen, A., Guerit, J.-M., Hinrichs, H.,
Ikeda, A., Luccas, F. J. C., & Rappelsberger, P. (1999). IFCN standards for digital recording
of clinical EEG. In G. Deuschl & A. Eisen (Eds.), Recommendations for the practice of
clinical neurophysiology: Guidelines of the international federation of clinical physiology,
Electroencephalography and clinical neurophysiology supplement (Vol. 52, 2nd ed.,
pp. 11–14). Amsterdam: Elsevier.
22. Wilson, S. B., & Emerson, R. (2002). Spike detection: A review and comparison of algo-
rithms. Clinical Neurophysiology, 113(12), 1873–1881.
23. Cohen, M. X. (2014). Analyzing neural time series data: Theory and practice. Cambridge,
MA: MIT Press.
24. Delorme, A., & Makeig, S. (2004). EEGLAB: An open source toolbox for analysis of single-
trial EEG dynamics including independent component analysis. Journal of Neuroscience
Methods, 134(1), 9–21.
25. Gwin, J. T., Gramann, K., Makeig, S., & Ferris, D. P. (2011). Electrocortical activity is
coupled to gait cycle phase during treadmill walking. NeuroImage, 54(2), 1289–1296.
26. Wagner, J., Solis-Escalante, T., Grieshofer, P., Neuper, C., Muller-Putz, G., & Scherer,
R. (2012). Level of participation in robotic-assisted treadmill walking modulates midline
sensorimotor EEG rhythms in able-bodied subjects. NeuroImage, 63(3), 1203–1211.
27. Kohli, S., & Casson, A. J. (2015). Towards out-of-the-lab EEG in uncontrolled environments:
Feasibility study of dry EEG recordings during exercise bike riding. IEEE Engineering in
Medicine and Biology Society, 2015, 1025–1028.
28. Zink, R., Hunyadi, B., Van Huffel, S., & De Vos, M. (2016). Mobile EEG on the bike:
Disentangling attentional and physical contributions to auditory attention tasks. Journal of
Neural Engineering, 13(4), 046017.
29. Mijovic, B., De Vos, M., Gligorijevic, I., Taelman, J., & Van Huffel, S. (2010). Source
separation from single-channel recordings by combining empirical-mode decomposition and
independent component analysis. IEEE Transactions on Biomedical Engineering, 57(9),
2188–2196.
30. Logesparan, L., Casson, A. J., & Rodriguez-Villegas, E. (2012). Optimal features for online
seizure detection. Medical & Biological Engineering & Computing, 50(7), 659–669.
31. Micheloyannis, S., Flitzanis, N., Papanikolaou, E., Bourkas, M., Terzakis, D., Arvanitis, S., &
Stam, C. J. (1998). Usefulness of non-linear EEG analysis. Acta Neurologica Scandinavica,
97(1), 13–19.
32. Mallat, S. (1999). A wavelet tour of signal processing (2nd ed.). San Diego: Academic.
33. Jentzsch, I., & Sommer, W. (2001). Sequence-sensitive subcomponents of P300: Topograph-
ical analyses and dipole source localization. Psychophysiology, 38(4), 607–621.
34. Ramoser, H., Muller-Gerking, J., & Pfurtscheller, G. (2000). Optimal spatial filtering of
single trial EEG during imagined hand movement. IEEE Transactions on Rehabilitation
Engineering, 8(4), 441–446.
35. Townsend, G., Graimann, B., & Pfurtscheller, G. (2004). Continuous EEG classification
during motor imagery-simulation of an asynchronous BCI. IEEE Transactions on Neural
Systems and Rehabilitation Engineering, 12(2), 258–265.
36. LaFleur, K., Cassady, K., Doud, A., Shades, K., Rogin, E., & He, B. (2013). Quadcopter
control in three-dimensional space using a noninvasive motor imagery-based brain–computer
interface. Journal of Neural Engineering, 10(4), 046003.
37. Gotman, J., & Gloor, P. (1976). Automatic recognition and quantification of interictal
epileptic activity in the human scalp EEG. Electroencephalography and Clinical Neurophys-
iology, 41(5), 513–529.
38. Pollock, V. E., Schneider, L. S., & Lyness, S. A. (1990). EEG amplitudes in healthy, late-
middle-aged and elderly adults: Normality of the distributions and correlations with age.
Electroencephalography and Clinical Neurophysiology, 75(4), 276–288.
39. Casson, A. J., & Rodriguez-Villegas, E. (2011). Interfacing biology and circuits: Quantifi-
cation and performance metrics. In K. Iniewski (Ed.), CMOS biomicrosystems: Where
electronics meet biology (pp. 3–32). Hoboken: Wiley.
40. Christensen, J. C., Estepp, J. R., Wilson, G. F., & Russell, C. A. (2011). The effects of day-to-
day variability of physiological data on operator functional state classification. NeuroImage,
59(1), 57–63.
41. Tallgren, P., Vanhatalo, S., Kaila, K., & Voipio, J. (2005). Evaluation of commercially
available electrodes and gels for recording of slow EEG potentials. Clinical Neurophysiol-
ogy, 116(4), 799–806.
42. Neuman, M. R. (2000). Biopotential electrodes. In J. D. Bronzino (Ed.), The biomedical
engineering handbook (2nd ed.). Boca Raton: CRC Press.
43. Huigen, E., Peper, A., & Grimbergen, C. A. (2002). Investigation into the origin of the noise
of surface electrodes. Medical & Biological Engineering & Computing, 40(3), 332–338.
44. Xu, J., Yazicioglu, R. F., Grundlehner, B., Harpe, P., Makinwa, K. A. A., & Van Hoof,
C. (2011). A 160 μW 8-channel active electrode system for EEG monitoring. IEEE Trans-
actions on Biomedical Circuits and System, 5(6), 555–567.
45. Ferree, T. C., Luu, P., Russell, G. S., & Tucker, D. M. (2001). Scalp electrode impedance,
infection risk, and EEG data quality. Clinical Neurophysiology, 112(3), 536–544.
46. Krachunov, S., & Casson, A. J. (2016). 3D printed dry EEG electrodes. Sensors, 16(10),
1635.
47. Taheri, B. A., Knight, R. T., & Smith, R. L. (1994). A dry electrode for EEG recording.
Electroencephalography and Clinical Neurophysiology, 90(5), 376–383.
48. Chi, Y. M., Jung, T. P., & Cauwenberghs, G. (2010). Dry-contact and noncontact biopotential
electrodes: Methodological review. IEEE Reviews in Biomedical Engineering, 3(1),
106–119.
49. Casson, A. J. (2016, August). An introduction to next generation EEG electrodes. IEEE
EMBC. Orlando: IEEE.
50. Lopez-Gordo, M. A., Sanchez-Morillo, D., & Pelayo Valle, F. (2014). Dry EEG electrodes.
Sensors, 14(7), 12847–12870.
51. Grass Technologies. http://www.grasstechnologies.com/. Accessed Jan 2017.
52. Debener, S., Emkes, R., De Vos, M., & Bleichner, M. (2015). Unobtrusive ambulatory EEG
using a smartphone and flexible printed electrodes around the ear. Scientific Reports, 5
(16743), 1–11.
53. Smith, P. E. M., & Wallace, S. J. (2001). Clinicians’ guide to epilepsy. London: Arnold.
54. Waterhouse, E. (2003). New horizons in ambulatory electroencephalography. IEEE Engi-
neering in Medicine and Biology Magazine, 22(3), 74–80.
55. Smith, S. J. M. (2005). EEG in the diagnosis, classification, and management of patients with
epilepsy. Journal of Neurology, Neurosurgery, and Psychiatry, 76(2), ii2–ii7.
56. Ebersole, J. S., & Bridgers, S. L. (1985). Direct comparison of 3- and 8-channel ambulatory
cassette EEG with intensive inpatient monitoring. Neurology, 35(6), 846–854.
57. Casson, A. J., Yates, D. C., Smith, S. J. M., Duncan, J. S., & Rodriguez-Villegas, E. (2010).
Wearable electroencephalography. IEEE Engineering in Medicine and Biology Magazine, 29
(3), 44–56.
58. Emotiv. https://www.emotiv.com/. Accessed Jan 2017.
59. Muse. http://www.choosemuse.com/. Accessed Jan 2017.
60. Neurosky. http://neurosky.com/. Accessed Jan 2017.
61. Rythm. https://rythm.co/. Accessed Jan 2017.
62. Kokoon. https://kokoon.io/. Accessed Jan 2017.
63. Badcock, N. A., Mousikou, P., Mahajan, Y., De Lissa, P., Thie, J., & McArthur, G. (2013).
Validation of the Emotiv EPOC (R) EEG gaming system for measuring research quality
auditory ERPs. PeerJ, 19(1), e38.
64. OpenBCI. http://openbci.com/. Accessed Jan 2017.
65. Mihajlovic, V., Grundlehner, B., Vullers, R., & Penders, J. (2015). Wearable, wireless EEG
solutions in daily life applications: What are we missing? IEEE Journal of Biomedical and
Health Informatics, 19(1), 6–21.
66. Lin, C. T., Liao, L. D., Liu, Y. H., Wang, I. J., Lin, B. S., & Chang, J. Y. (2011). Novel dry
polymer foam electrodes for long-term EEG measurement. IEEE Transactions on Biomedical
Engineering, 58(5), 1200–1207.
67. Looney, D., Kidmose, P., Park, C., Ungstrup, M., Rank, M. L., Rosenkranz, K., & Mandic,
D. (2012). The in-the-ear recording concept: User-centered and wearable brain monitoring.
IEEE Pulse, 3(6), 32–42.
68. Kidmose, P., Looney, D., Ungstrup, M., Rank, M. L., & Mandic, D. P. (2013). A study of
evoked potentials from ear-EEG. IEEE Transactions on Biomedical Engineering, 60(10),
2824–2830.
69. Kim, D.-H., Lu, N., Ma, R., Kim, Y.-S., Kim, R.-H., Wang, S., Wu, J., Won, S. M., Tao, H.,
Islam, A., Yu, K. J., Kim, T., Chowdhury, R., Ying, M., Xu, L., Li, M., Chung, H.-J., Keum,
H., McCormick, M., Liu, P., Zhang, Y.-W., Omenetto, F. G., Huang, Y., Coleman, T., &
Rogers, J. A. (2011). Epidermal electronics. Science, 333(6044), 838–843.
70. Norton, J. J., Lee, D. S., Lee, J. W., Lee, W., Kwon, O., Won, P., Jung, S. Y., Cheng, H.,
Jeong, J. W., Akce, A., Umunna, S., Na, I., Kwon, Y. H., Wang, X. Q., Liu, Z., Paik, U.,
Huang, Y., Bretl, T., Yeo, W. H., & Rogers, J. A. (2015). Soft, curved electrode systems
capable of integration on the auricle as a persistent brain-computer interface. Proceedings of
the National Academy of Sciences of the United States of America, 112(13), 3920–3925.
71. Batchelor, J. C., Yeates, S. G., & Casson, A. J. (2016). Conformal electronics for longitudinal
bio-sensing in at-home assistive and rehabilitative devices. IEEE Engineering in Medicine
and Biology Society, 2016, 3159–3162.
72. Sanchez-Romaguera, V., Ziai, M. A., Oyeka, D., Barbosa, S., Wheeler, J. S. R., Batchelor,
J. C., Parker, E. A., & Yeates, S. G. (2013). Towards inkjet-printed low cost passive UHF
RFID skin mounted tattoo paper tags based on silver nanoparticle inks. Journal of Materials
Chemistry C, 1(39), 6395–6402.
73. Ziai, M. A., & Batchelor, J. C. (2011). Temporary on-skin passive UHF RFID transfer tag.
IEEE Transactions on Antennas and Propagation, 59(10), 3565–3571.
74. Iber, C., Ancoli-Israel, S., Chesson, A., & Quan, S. F. (2007). The AASM manual for the
scoring of sleep and associated events: Rules, terminology and technical specifications.
Westchester: American Academy of Sleep Medicine.
75. Neligan, A., & Sander, J. W. (2015). The incidence and prevalence of epilepsy. Available
https://www.epilepsysociety.org.uk/. Accessed Jan 2017.
76. Browne, T. R., & Holmes, G. L. (2001). Epilepsy. The New England Journal of Medicine,
344(15), 1145–1151.
77. Epilepsy society, diagnosing epilepsy. Available https://www.epilepsysociety.org.uk.
Accessed Jan 2017.
78. National Institute for Clinical Excellence. (2004). NICE guidelines: The diagnosis and
management of the epilepsies in adults and children in primary and secondary care.
London: NICE.
79. Rechtschaffen, A., & Kales, A. (Eds.). (1968). A manual of standardized terminology,
techniques and scoring system for sleep stages of human subjects. Washington, DC: Public
Health Service, U.S. Government Printing Office.
80. Carney, P. R., Berry, R. B., & Geyer, J. D. (Eds.). (2005). Clinical sleep disorders. Phila-
delphia: Lippincott Williams and Wilkins.
81. Colten, H. R., & Altevogt, B. M. (Eds.). (2006). Sleep disorders and sleep deprivation: An
unmet public health problem. Washington, DC: National Academies Press.
82. Nicolas-Alonso, L. F., & Gomez-Gil, J. (2012). Brain computer interfaces, a review. Sensors,
12(2), 1211–1279.
83. Ramos-Murguialday, A., Broetz, D., Rea, M., Laer, L., Yilmaz, O., Brasil, F. L., Liberati, G.,
Curado, M. R., Garcia-Cossio, E., Vyziotis, A., Cho, W., Agostini, M., Soares, E., Soekadar,
S., Caria, A., Cohen, L. G., & Birbaumer, N. (2013). Brain-machine-interface in chronic
stroke rehabilitation: A controlled study. Annals of Neurology, 74(1), 100–108.
84. Neto, E., Allen, E. A., Aurlien, H., Nordby, H., & Eichele, T. (2015). EEG spectral features
discriminate between Alzheimer’s and vascular dementia. Frontiers in Neurology, 6(25),
1–9.
85. Wolpaw, J. R., McFarland, D. J., Neat, G. W., & Forneris, C. A. (1991). An EEG-based brain-
computer interface for cursor control. Electroencephalography and Clinical Neurophysiol-
ogy, 78(3), 252–259.
86. Zander, T. O., & Kothe, C. (2011). Towards passive brain computer interfaces: Applying
brain computer interface technology to human machine systems in general. Journal of Neural
Engineering, 8(2), 025005.
87. Carlson, T., & Millan, J. R. (2013). Brain-controlled wheelchairs: A robotic architecture.
IEEE Journal of Robotics and Automation, 20(1), 65–73.
88. Kubler, A., Mushahwar, V. K., Hochberg, L. R., & Donoghue, J. P. (2004). BCI meeting
2005—Workshop on clinical issues and applications. IEEE Transactions on Neural Systems
and Rehabilitation Engineering, 14(2), 131–134.
89. Chen, X., Wang, Y., Nakanishi, M., Gao, X., Jung, T.-P., & Gao, S. (2015). High-speed
spelling with a noninvasive brain–computer interface. Proceedings of the National Academy
of Sciences of the United States of America, 112(44), 6058–6067.
90. Guger, C., Daban, S., Sellers, E., Holzner, C., Krausz, G., Carabalona, R., Gramatica, F., &
Edlinger, G. (2009). How many people are able to control a P300-based brain–computer
interface (BCI)? Neuroscience Letters, 462(1), 94–98.
91. Ekandem, J. I., Davis, T. A., Alvarez, I., James, M. T., & Gilbert, J. E. (2012). Evaluating the
ergonomics of BCI devices for research and experimentation. Ergonomics, 55(5), 592–598.
92. Dijksterhuis, C., De Waard, D., Brookhuis, K., Mulder, B., & De Jong, R. (2013). Classifying
visuomotor workload in a driving simulator using subject specific spatial brain patterns.
Frontiers in Neuroscience, 393(7), 149.
93. Casson, A. J. (2014). Artificial Neural Network classification of operator workload with an
assessment of time variation and noise-enhancement to increase performance. Frontiers in
Neuroscience, 8(372), 1–10.
94. Wilson, G. F., & Russell, C. A. (2007). Performance enhancement in a UAV task using
psychophysiological determined adaptive aiding. Human Factors, 49(6), 1005–1019.
95. Ayaz, H., Shewokis, P. A., Bunce, S., Izzetoglu, K., Willems, B., & Onaral, B. (2012).
Optical brain monitoring for operator training and mental workload assessment. NeuroImage,
59(1), 36–47.
96. Transparency market research. Available http://www.prweb.com/releases/2013/11/
prweb11337791.htm. Accessed Jan 2017.
97. Surangsrirat, D., & Intarapanich, A. (2015, April). Analysis of the meditation brainwave from
consumer EEG device. IEEE SoutheastCon, Fort Lauderdale.
98. Lee, N., Broderick, A. J., & Chamberlain, L. (2007). What is “neuromarketing”? A discussion
and agenda for future research. International Journal of Psychophysiology, 63(2), 199–204.
99. Koelstra, S., Muehl, C., Soleymani, M., Lee, J.-S., Yazdani, A., Ebrahimi, T., Pun, T.,
Nijholt, A., & Patras, I. (2011). DEAP: A database for emotion analysis using physiological
signals. IEEE Transactions on Affective Computing, 3(1), 18–31.
100. Little, S., Pogosyan, A., Neal, S., Zavala, B., Zrinzo, L., Hariz, M., Foltynie, T., Limousin, P.,
Ashkan, K., FitzGerald, J., Green, A. L., Aziz, T. Z., & Brown, P. (2013). Adaptive deep
brain stimulation in advanced Parkinson disease. Annals of Neurology, 74(3), 449–457.
101. Stanslaski, S., Afshar, P., Cong, P., Giftakis, J., Stypulkowski, P., Carlson, D., Linde, D.,
Ullestad, D., Avestruz, A.-T., & Denison, T. (2012). Design and validation of a fully
implantable, chronic, closed-loop neuromodulation device with concurrent sensing and
stimulation. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 20(4),
410–421.
102. Famm, K., Litt, B., Tracey, K. J., Boyden, E. S., & Slaoui, M. (2013). Drug discovery: A
jump-start for electroceuticals. Nature, 496(7444), 159–161.
103. Paulus, W. (2011). Transcranial electrical stimulation (tES–tDCS; tRNS, tACS) methods.
Neuropsychological Rehabilitation, 21(5), 602–617.
104. Kohli, S., & Casson, A. J. (2015). Removal of transcranial ac current stimulation artifact from
simultaneous EEG recordings by superposition of moving averages. IEEE Engineering in
Medicine and Biology Society, 2015, 3436–3439.
105. Pfurtscheller, G., & Neuper, C. (2001). Motor imagery and direct brain-computer communi-
cation. Proceedings of the IEEE, 89(7), 1123–1134.
http://www.springer.com/978-3-319-69361-3
View publication stats

Final

Uploaded by

Copyright:

Available Formats

Final

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Final

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Chapter · January 2018

Alex Casson Meera Dulabh

SEE PROFILE SEE PROFILE

Gealtine Head Phantom View project

The user has requested enhancement of the downloaded file.

Alexander J. Casson, Mohammed Abdulaal, Meera Dulabh,

The electroencephalogram (EEG) is a widely used non-invasive method for mon-

A.J. Casson (*) • M. Abdulaal • S. Kohli

© Springer International Publishing AG 2018 45

2.2.2.1 Origin of EEG

2.2.2.2 Typical Signals

“transient[s], clearly distinguished from background activity . . . [with] a

• Slow oscillations: A clear sub 1 Hz oscillation with negative going state,

2.2.3.1 Basic Set Up

Typical signals detected by scalp-mounted electrodes are in the range 1–150 μV

in intensive care units [19]. Similarly, the Electrocorticogram (ECoG) [2] is

2.2.3.2 Practical Set Up

2.2.4 Signal Processing and Data Analysis

Fig. 2.8 Basic EEG signal

• To disambiguate analysis, giving a single justifiable interpretation of an EEG

2.2.4.2 Motion Artefact Removal

For other sections of data, it is possible to remove motion interference algorith-

2.2.4.3 Data-Driven Machine Learning

2.2.4.4 Performance Assessment

Finally, attention must be given to the performance assessment methodologies for

2.3 Modality of Measurement

the impedance of the electrode–electrolyte interface, and Re is the material resis-

Fig. 2.11 Active EEG

different papers are difficult to do accurately. A recent review comparing different

2.3.3 Electrode Attachments to the Scalp

2.3.4.2 Research Lab/Inpatient

2.3.4.5 Beyond Wearables

2.4.2 Medical, Clinical and Neuroscience Research

corresponds to approximately 96,000 people who require hospital-based treatment,

2.4.3 Brain–Computer Interfaces (BCIs)

The most widely developed non-directly medical applications of EEG technology

2.4.4 Consumer Neuroscience

2.5 Conclusions and Future Prospects

present transcranial stimulation introduces significant artefacts into simultaneous

Electroencephalography and clinical neurophysiology supplement (Vol. 52, 2nd ed.,

View publication stats

You might also like