Handbook of Clinical Neurology, Vol.

168 (3rd series)

Brain-Computer Interfaces
N.F. Ramsey and J. del R. Millán, Editors
https://doi.org/10.1016/B978-0-444-63934-9.00018-4
Copyright © 2020 Elsevier B.V. All rights reserved

Chapter 18

Electroencephalography
€
GERNOT R. MULLER-PUTZ*

Institute for Neural Engineering, Laboratory of Brain-Computer Interfaces, Graz University of Technology, Graz, Austria

Abstract
The electroencephalogram (EEG) was invented almost 100 years ago and is still a method of choice for
many research questions, even applications—from functional brain imaging in neuroscientific investiga-
tions during movement to real-time applications like brain-computer interfacing. This chapter gives some
background information on the establishment and properties of the EEG.
This chapter starts with a closer look at the sources of EEG at a micro or neuronal level, followed by
recording techniques, types of electrodes, and common EEG artifacts. Then an overview on EEG
phenomena, namely, spontaneous EEG and event-related potentials build the middle part of this chapter.
The last part discusses brain signals, which are used in current BCI research, including short descriptions
and examples of applications.

INTRODUCTION cause local extracellular potential changes. The superpo-

sition of these differences in potentials has been named
The electroencephalogram (EEG) has been used as a sci-
local field potential (LFP) (Buzsáki et al., 2012). The
entific tool for almost 100 years. Hans Berger, a German
frequency spectrum of LFPs is quite broad starting from
neurologist and psychiatrist, discovered a specific EEG
DC (direct current) up to several hundred Hertz. Synaptic
rhythm, called alpha oscillations in 1924 and published
transmissions and action potentials (APs) are regarded as
his findings in 1929 (Berger, 1929). main sources of LFPs (Einevoll et al., 2013). A plausible
EEG is one of the standard methods to measure brain
assumption is that synaptic transmissions are the source
activity in many fields and the main source of signals
of low frequency components and APs are the source of
for noninvasive brain-computer interfaces (BCIs). This
high frequency components (>500 Hz) (Buzsáki et al.,
chapter highlights the physiologic foundation for the
2012). If ohmic impedances (i.e., resistance independent
properties of the EEG, different recording techniques,
of frequencies) and electric dipole sources are assumed,
and their implications. The final section describes current the contribution of a single source to the LFP decays with
BCI research applications using different EEG signals. the square of the distance (Nunez and Srinivasan, 2006).
However, the type of impedance in brain tissue is contro-
GENERAL ESTABLISHMENT OF EEG versial. Typically, brain tissue is assumed to have high
pass (i.e., high frequencies pass, low frequencies are
From neuron to EEG
dampened) properties, but there are also indications for
The human brain is estimated to comprise 100 billion low pass properties (Grimnes and Martinsen, 2000;
neurons, with each neuron having approximately Bedard et al., 2004).
10,000 connections to other neurons. This huge, electri- A paper of Logothetis et al. (2007) showed that
cally active neuronal network can be divided into many pure ohmic impedance is a sufficient assumption for
subnetworks. Ionic currents present in these subnetworks low frequencies (<1000 Hz) (Logothetis et al., 2007).

*Correspondence to: Gernot R. M€uller-Putz, Institute for Neural Engineering, Laboratory of Brain-Computer Interfaces, Graz
University of Technology, Graz, Austria. Tel: +43-316-87330700, E-mail: [email protected]
250 €
G.R. MULLER-PUTZ
A recent modeling of the LFP conducted by Linden et al.
(2011) indicated that for uncorrelated activity, a reach of
200 mm is a realistic assumption. For correlated activity
this assessment is more difficult, but they concluded that
“…the LFP recorded by an electrode is dominated by
populations with substantial synaptic processes in the
recording layer” (Linden et al., 2011). Hence, nearby
sources contribute most to the LFP and contributions
from distant sources are subject to strong attenuation.
As a result of this, it is only possible to measure the
collective activity of a large number of neurons at the
scalp (Fabiani et al., 2007).
EEG is by far the most common noninvasive method
for measuring electrical brain activity. It can measure the
brain potentials through various types of electrodes (see
Section “Electrode principles to measure EEG”), which Fig. 18.1. Sketch of BCI signal sources. I–VI mark the cortical
get mounted at the scalp level with the help of an EEG layers. Pyramidal cells in cortical layers 5 and 6 are highlighted
cap. These measured scalp potentials are a modified ver- in green. Their apical and basal synapses are color coded. The
sion of the LFPs (Buzsáki et al., 2012). The modification spatial and temporal dendritic integration of synaptic transmis-
has at least two causes. First, as described earlier, the sion leads to formation of dipoles. If millions of neurons
receive synchronous basal or apical synaptic transmissions,
electric field decays with the square of the distance from
the resulting electrical field propagates over large distances
the source, and therefore, the LFP substantially attenu-
and is even detectable at the scalp where it is called EEG. Mod-
ates when it reaches the scalp electrodes. Second, volume ified from Steyrl D., Kobler R.J. and M€ uller-Putz G.R., 2016.
conductance of the head’s tissues (mainly brain, cerebral On similarities and differences of invasive and non-invasive
fluid, skull, and scalp) causes spatial smoothing over an electrical brain signals in brain-computer interfacing.
area of about 10 cm2 (Buzsáki et al., 2012). J Biomed Sci Eng 9: 393–398.
Due to the attenuation and smoothing, only synchro-
nous brain activity (i.e., brain activity that sums up over The larger the synchronously active cell population,
brain areas) can be measured at the scalp level. Rhythms the higher the potential deflection, i.e., amplitude in
that occur synchronously are typical for lower frequency the EEG. In contrast, in deep brain structures (e.g., amyg-
ranges of the LFP. The low frequency components of dala), neurons’ electric fields are typically oriented in dif-
LFPs are mainly caused by correlated synaptic transmis- ferent directions, which impedes the summation process
sions and can be seen as neural dipoles in parallel pyra- (i.e., they cancel each other out) (Lorente de No, 1947).
mid cells (see Fig. 18.1) (Buzsáki et al., 2012; Einevoll Thus, such structures do not generate large summated
et al., 2013). Since only afferent APs lead to synaptic dipoles (Harmon-Jones and Beer, 2012) and, therefore,
transmissions, it can also be assumed that the major con- cannot be assessed by EEG oscillations at the scalp. This
tribution to LFP at lower frequencies comes from these also holds to some extent for the orientation in sulci.
afferent APs in cortex layers 1–4 (see Fig. 18.1). Considering EEG’s physiologic foundations that have
APs cause synaptic transmissions, but information been outlined so far, an important implication for correct
coded in AP spike trains is not one-to-one equivalent interpretation of EEG signals emerges. EEG oscillations
to information in low frequency components of the recorded at the scalp only represent a subset of the elec-
LFP (Hodgkin and Huxley, 1990; Einevoll et al., trical brain activity at a particular point in time. Research
2013). In fact, the connection from APs via synaptic indicates that low-frequency oscillations (e.g., theta)
transmissions to the LFPs is not entirely understood span larger neural populations, while higher-frequency
yet. Ionic transmembrane currents can be well described oscillations (e.g., gamma) span smaller neural assem-
by models; however, our understanding is limited by blies (Buzsáki and Draguhn, 2004).
influences like the feedback of the LFP to surrounding Gevins and Smith outlined five major determinants of
cell activity and other effects (Hodgkin and Huxley, the degree to which potentials arising in the cortex are
1952; Goldman, 2004; Einevoll et al., 2013). measureable at scalp level (Gevins and Smith, 2006):
In summary, APs of afferent fibers in the cortex can (1) signal amplitude at the cortex, (2) size of the region
cause synaptic transmissions. Correlated synaptic trans- over which postsynaptic potentials occur synchronously,
missions form parallel neural dipoles, which contribute (3) proportion of cells that are in synchrony in that
most to synchronous low frequency components of the region, (4) location and orientation of the activated cor-
LFP and, therefore, also contribute to the scalp EEG. tical region in relation to the surface of the scalp, and (5),
 ELECTROENCEPHALOGRAPHY 251
the amount of signal dampening and spatial smearing sites on the left hemisphere and even numbers
generated by conduction through the liquor, skull, and indicate sites on the right hemisphere. Midline elec-
other tissue layers. trodes (i.e., on the virtual line connecting the nasion
and the inion, where the vertex corresponds to its
RECORDING, ELECTRODES, half-length) have “z” as indicator. Moreover, num-
AMPLIFIERS, AND ARTIFACTS bers increase as distance from the midline increases
(see, for example, Fz, F3, F7 in Fig. 18.2).
Electrode positioning system
However, for specific BCI applications, researchers
In EEG recordings, electrode locations are based on stan- often choose individualized and end-user specific elec-
dard position systems. The 10–20 system originally pro- trode positions. Still, the general rules described here
posed by Jasper (1958) is one of the most internationally are usually adopted.
recognized methods to describe the locations of the EEG
scalp electrodes and it ensures that the interelectrode dis-
tances are equal. Here, electrodes are placed at sites 10% Electrode principles to measure EEG
and 20% from four anatomical landmarks: the nasion,
At the very beginning, in 1924, scientists inserted steel
inion, left, and right preauricular points. However, to
needles into the subcutaneous tissue of the scalp and used
achieve a higher spatial resolution, extra electrodes can
galvanometers to visualize and interpret the recorded sig-
be added to the 10–20 system, leading to more detailed
nals (Berger, 1929). The quality and the interpretability
systems such as the 10–10 or 10–5 systems. For that,
of the signals improved with technological developments
intermediate positions have been added between those
to amplify the very small signals. Still a standard nowa-
of the original 10–20 system (Oostenveld and
days, silver chloride (AgCl) covered electrodes were
Praamstra, 2001). Fig. 18.2 shows a 10–5 system where
introduced by Berger in 1931 (Collura, 1993).
only the original 10–20 system electrodes are labeled.
When measuring EEG, a conductive connection to
The existing naming convention for electrode positions
bridge the gap between the electrode and the skin
is shown in Fig. 18.2. The following rules apply:
surface has to be introduced. Currently, there are three
(1) The first character refers to the cortical area common types of electrodes: gel-based, water-based,
(F ¼ frontal area, C ¼ central area, P ¼ parietal area, or dry-electrodes. The latter, as the name indicates, does
T ¼ temporal area, and O ¼ occipital area). Elec- not need an additional conductive substance. Fig. 18.3
trodes between these areas are labeled using two shows different types of EEG electrodes.
characters (e.g., FC ¼ frontal–central). Gel-based electrodes can be subdivided based on
(2) A number (e.g., P3) or another character (e.g., Cz) the usage of abrasive gel or hydrogel. Abrasive gel is
follows after the first letter. Odd numbers indicate mainly used in combination with passive electrodes

Fig. 18.2. Scheme of a 10–5 electrode system, based on Oostenveld and Praamstra (2001). Selected electrode positions are shown.
A1 is the earlobe, P shows the preauricular point. It is on the line between the Nasion and Inion above the tragus.
252 €
G.R. MULLER-PUTZ

Fig. 18.3. Examples of electrodes. (A) Old cup electrode, (B) old passive sintered AgCl electrode, (C) gel-based passive Ag/AgCl
ring electrode (from EasyCap), (D) gel-based active Ag/AgCl electrode (g.LADYbird from g.tec), (E) gel-based active Ag/AgCl
(actiCAP, BrainProducts), (F) passive dry electrode with gold-coated pins (g.SAHARA electrode from g.tec), (G) (tap) water-
based passive electrode (Mobita, TMSi), (H) (tap) water-based passive electrode (BitBrain Technologies), (I) passive dry elec-
trodes with pins (BitBrain Technologies).

(i.e., direct connection between the electrode and the When looking at electrode technology from a BCI
amplifier input). In contrast, the hydrogel is used for end-user perspective, comfort should be maximized
active electrodes. On active electrodes, a tiny preampli- and extra inconveniences eliminated (e.g., washing the
fier sits on the electrode and increases the robustness of hair). From a technical point of view, the signal quality
the signals before being conducted to the main amplifier. has to be optimal to make the BCI perform effectively
The main difference between these two types of gels is and efficiently. A system, which is user-friendly and
that with the abrasive gel, the topmost layer of the skin, provides, at the same time, the necessary signal quality
consisting of dead cells and a small amount of fat, has to is therefore challenging to develop. A recent work
be removed in a time-consuming procedure to decrease (Pinegger et al., 2016) evaluated three different commer-
the impedance. This can lead to skin irritation, infection, cially available EEG acquisition systems. They differed
or inflammation. For both types of gels, it is necessary for in the type of electrodes (gel-, water-, or dry-based), the
the participants to wash their hair after the measurement. amplifier technique, and the data transmission method.
Water-based electrodes use a felt or other fabric material Every system was tested with regard to three different
soaked in water or saline solution to connect the electrode aspects, namely, (i) technical, (ii) BCI effectiveness
with the skin. Using tap water-soaked fabric to connect and efficiency (P300 for communication and control),
the two surfaces is a relatively new and practical method. and (iii) user satisfaction (comfort). The findings indicate
This type of electrode should deliver a very good signal that the water-based system had the lowest short circuit
quality, the setup is less time-consuming, and no hair noise level, the gel-based system had the highest P300
wash is needed after the measurement (Volosyak et al., spelling accuracies, and the dry electrode-based system
2010; Pinegger et al., 2016). caused the least inconveniences for the user (Pinegger
Dry electrodes, in contrast, work without any conduc- et al., 2016).
tive substance. Pins made of metal alloy or conductive Another recent study (Melnik et al., 2017) investi-
rubber are pressed directly onto the skin, and rely on gated the variance across different EEG systems com-
small amounts of existing perspiration to get connected pared to the variance across subjects or sessions. The
to the skin. Several studies have highlighted the advan- authors tested four different systems, one mobile EEG
tages of different dry electrode-based systems (Zander system with dry electrodes, one affordable system with
et al., 2011; Guger et al., 2012; Mota et al., 2013). How- a low number of channels, and two standard gel-based
ever, experience shows that one main disadvantage of research-grade systems. They recorded four subjects
this type of electrode is their sensitivity to movement three times with each of the four EEG systems in six
artifacts. different standard EEG paradigms. The authors describe
 ELECTROENCEPHALOGRAPHY 253
that the two standard research EEG systems had no Praamstra, 2001; Scherer et al., 2007) or removal
significantly different means from each other across all (Schl€ogl et al., 2007 ; Daly et al., 2015) are important
paradigms. However, the two other EEG systems dem- to record correct data (i.e., non-artifactual features)
onstrated different mean values from one or both of and, thereby, obtain reliable classification results.
the two standard research-grade EEG systems in at least
half of the paradigms.
BRAIN SIGNALS AND THEIR USE IN BCI
It can be concluded that the type of application and its
requirements are important to decide which electrode EEG research distinguishes between two major types of
technology should be used (Nijboer et al., 2015). Of brain activity. Consequently, there are also two major
course, this is often strongly coupled with the choice types of noninvasive and EEG-based types of BCIs:
of the amplifier, since many of those electrodes are (i) spontaneous EEG (also referred to as endogenous
company-specific. Nowadays, all amplifiers have built- or continuous EEG), which is based on internally
in analog-to-digital conversion and get connected to induced processes and mental tasks that mainly generate
the computer via USB or network connection. changes in the ongoing EEG and (ii) event-related poten-
tials, which are based on events or external stimuli.
In this section these two different phenomena are dis-
EEG artifacts
cussed and prominent BCI examples are given thereafter.
When doing BCI research and single-trial classification it
is of great importance to process clean EEG data.
Spontaneous EEG
However, there is always the danger of having contam-
inated EEG signals and therefore BCI researchers must The spontaneous or continuous EEG is the measurable
consider artifacts carefully. Generally, technical and part of brain activity that goes on permanently in the
biologic artifacts exist. living individual. In the healthy waking brain, the
The main sources of technical artifacts are primarily peak-to-peak amplitude of this signal is typically under
external electrical and electromagnetic noise coming 75 mV but sometimes increases to 100 mV (Gevins and
from power lines, electric lights, or other fields. Poor Smith, 2006). A considerable portion of the signal power
contact can lead to high impedances and thus foster originates from rhythmic oscillations in a frequency
electromagnetic artifacts. Wrong electrode material can bandwidth from below 1 Hz to approximately 40 Hz,
lead to high pass effects that can hide the requested even though higher frequencies are also measureable
signal. In addition, mainly hidden to the BCI researcher up to 100 Hz (Schomer and da Silva, 2012). This wide
are the amplifier noise and quantization noise of the frequency range is subdivided into smaller, functional
analogue-to-digital conversion. Aliasing effects due to ranges with associated names (Schomer and da
wrongly adjusted filters can be problematic. Experience Silva, 2012).
shows that major countermeasures for technical artifacts The alpha rhythm is characterized by medium-
include: shielding the recording system, using filters frequency activity (8–13 Hz) and generally indicates
(e.g., notch filters to remove power line noise), and high states of relaxed wakefulness in healthy adults (Berger,
quality amplifiers. A proper grounding of the participants 1929). The amplitude of these oscillations is typically
to reach potential equalization between participant and very large and can be up to several tens of mV. This wave
measurement system is mandatory. type is also common during resting periods in which
The main sources of biologic artifacts are partici- people have their eyes closed, then amplitudes are largest
pants’ muscle—electromyogram (EMG)—activities in the occipital areas. Based on this finding, researchers
(e.g., neck, face), eye blinks, and eye movements. Slight have argued that alpha waves constitute a neural corre-
baseline drifts (drift of the zeroline of the signal) due to late of cognitive inactivity, also referred to as cortical
sweating can also be problematic. While EMG occurs in “idling” (Pfurtscheller et al., 1996). However, studies
a range between 20 and 1500 Hz, the electrooculogram with evoked EEG activity (i.e., ERP investigations) have
(EOG) covers a narrow low frequency range from DC found that alpha rhythms may indicate different forms of
up to 10 Hz. Depending on the type of BCI study or information processing in which different alpha
application, concurrent recording of the EOG and subbands (e.g., 8–10 and 10–13 Hz) are dedicated to dif-
EMG are advised for applying detection or artifact ferent functional processes (Niedermeyer, 1997;
removal algorithms. Artifacts must be detected and Klimesch, 1999). Alpha rhythms originating from senso-
somehow indicated in an online system, where EEG data rimotor areas are also known as mu rhythms and can be
is processed and feedback to a user is generated. Depend- further subdivided into lower and higher mu rhythm
ing on the type of processing—offline or online—both (Pfurtscheller et al., 2000). Large amplitudes indicate
visual and automatic artifact detection (Oostenveld and resting sensorimotor areas.
254 €
G.R. MULLER-PUTZ
Beta oscillations are characterized by medium- to synchronization (ERD(S)) method introduced by
high-frequency activity (13–30 Hz) related to various Pfurtscheller in the late 1970s (Pfurtscheller and
mental states, such as active concentration, task engage- Aranibar, 1977) and described in detail in Pfurtscheller
ment, excitement, anxiety, attention, or vigilance. It is and Lopes da Silva (1999), compares band power values
also a marker for sensorimotor activity. The amplitudes during an activity period with the band power of a refer-
of this wave are usually in the mV. Beta activity primarily ence (resting) period. This comparison results in relative
constitutes an excitatory mechanism (Pfurtscheller and band power changes given in percentages. By the calcu-
Lopes da Silva, 1999) (see Fig. 18.4E). lation of these values for several frequency bands, the
Gamma oscillations are characterized by very ERD(S) maps represented as time-frequency maps can
high-frequency activity (30–200 Hz, but typically not be obtained (Graimann et al., 2002) (see Fig. 18.4D).
measurable by EEG when higher than 100 Hz). These There are many other methods available that exploit
oscillations are often associated with arousal and percep- these EEG oscillatory components either for direct
tual binding mechanisms (i.e., integration of various feedback (neuro-feedback) (Birbaumer et al., 1999) or
aspects of a stimulus into a coherent overall perception). machine learning based BCIs. The most common are
The amplitudes are already rather small and usually band power (Guger et al., 2000), common spatial pat-
between 1 and 2 mV (Hughes, 2008). terns (Ramoser et al., 2000; Blankertz et al., 2008),
Delta waves are characterized by very low-frequency and many others (Wolpaw and Wolpaw, 2012).
activity (below 1–4 Hz), which usually relates to deep Of high importance, though, is how these oscillatory
and unconscious sleep in healthy humans. Delta waves components can be mentally changed or influenced by
(amplitude can be several tenths of mV) are also associ- study participants or end-users.
ated with pathologic neural states, such as coma or the
loss of consciousness. Generally, delta activity dimin-
MOTOR IMAGERY AND OTHER MENTAL TASKS
ishes with increasing age, which suggests that delta
activity is primarily an inhibitory mechanism (Hobson One of the first mental strategies was to imagine hand
and Pace-Schott, 2002). movements (motor imagery, MI) (Pfurtscheller and
Theta waves occur as low-frequency activity (4–8 Hz) Neuper, 1997; Cincotti et al., 2003; Munzert et al.,
and are typically associated with specific sleep states, 2009). MI describes the mental rehearsal of a motor task
drowsiness, and meditation. However, in the literature, without its execution. Typical kinesthetic MI tasks are:
frontal midline theta is also described. This type has been (1) sustained imagination of squeezing a training ball
associated with mental effort, suggesting that attention is (Kaiser et al., 2014; Coyle et al., 2015), (2) repetitive
directed to an existing stimulus. In general, the amplitude opening and closing of the hand (Ramoser et al.,
of theta waves is typically between 8 and 10 mV (Cahn 2000; Pfurtscheller et al., 2003), or (3) sustained/
and Polich, 2006). repeated movement imagination of both feet, e.g.,
It is important to mention that these rhythms can be dorsi- or plantar-flexion of both feet (M€uller-Putz
time-locked to an event; however, they are always et al., 2007; Hashimoto and Ushiba, 2013). Further-
nonphase-locked. This means, when one compares more, this phenomenon can be triggered by an external
similar trials the amplitude behavior may be similar, event, and users may induce it by actively performing
however, the oscillations of the EEG rhythms will not the designated task (Pfurtscheller et al., 1997; Mason
have the same phase. Therefore, with simple averaging and Birch, 2000; Millán and Mourino, 2003). This fact
techniques, it is not possible to extract meaningful is used in noncue-guided, asynchronous BCI scenarios,
information from the signals. The main challenge in where users decide for themselves when to establish
BCI-research though is, to identify oscillatory patterns in control (Scherer et al., 2004; M€uller-Putz et al.,
EEG without knowing the time point of establishment— 2005a,b; Leeb et al., 2007).
i.e., in an asynchronous BCI application an end user Motor imagery can be an efficient strategy for control-
may wish to use the system without any external pace, ling a BCI based on the modulation of rhythms of the
stimulus, or cue. sensorimotor cortex also known as SMR-based BCI
There are many methods to analyze spontaneous (Scherer et al., 2008; Neuper et al., 2009; Blankertz
EEG, but in contrast to ERP analysis, the simple calcu- et al., 2010; Pichiorri et al., 2011; Faller et al., 2012;
lation of an average over trials does not work (see Kreilinger et al., 2013). The SMR-BCI uses the power
Section “Event-related potentials”). To get a first and decreases/increases as a feature for discriminating
solid impression of spontaneous EEG activity related between two or more different MIs.
to various conditions, one of the standard methods used Beside motor imagination, there are also other
in BCI research is to compare power values with respect mental tasks that can be used and lead to successful
to a reference period. The so-called event-related (de) modulation of EEG patterns (Harmony et al., 1996;
 ELECTROENCEPHALOGRAPHY 255

Fig. 18.4. Various EEG signals and patterns. (A) P300 target signal (blue) and average nontarget responses (red). (B) MRCP from
10 subjects averaged over 1000 trials (Sburlea et al., 2015b). (C) SSVEP spectra of focused attention to four different flashing
lights: 6, 7, 8, 13 Hz (M€uller-Putz et al., 2005a). (D) ERD map of a Laplacian derivation of Cz in an end-user with spinal cord
injury during foot motor imagery. (E) Examples of single EEG trials, Laplacian derivation, from Cz corresponding to (D).
256 €
G.R. MULLER-PUTZ
Obermaier et al., 2001; Millán et al., 2002; Cabrera et al., ERP signals are typically not very strong, and, hence,
2010; Jeunet et al., 2016). For example, distinct levels of it is difficult to distinguish them from the spontaneous
kinesthetic attention (from focused attention to mind wan- EEG in raw, single-trial, data. However, by repeated pre-
dering) in a continuous passive mobilization task have sentation of stimuli and averaging of the EEG responses,
been shown to have modulatory effects at the level of these ERPs can be made visible. Since ongoing EEG is
theta, alpha, and beta frequency bands (Melinscak et al., not time-locked and not phase-locked to the stimulus,
2016). Another study (Friedrich et al., 2013) explored averaging increases the signal-to-noise ratio (Regan,
seven mental tasks (mental rotation, word association, 1989; Fabiani et al., 2007; Luck, 2014). In BCI research,
auditory imagery, mental subtraction, spatial navigation, however, our goal is usually to achieve single-trial detec-
left hand MI, both feet MI) in a user-centered approach, tion, in which case a specific paradigm design, signal
and the best four classes were used to build a BCI. Later processing techniques, and machine learning approaches
on, this approach was successfully used to apply a BCI in are necessary (see Chapters 2 and 23).
end-users with disabilities (Scherer et al., 2015). Various types of BCIs based on different ERPs as well
as specific ERP-components are discussed next.
Event-related potentials
Event-related potentials (ERPs) mainly originate from
specific external stimuli. Regan (1989) gives the follow- P300 COMPONENT
ing definition:
The P300 is regarded as indicator of information pro-
“An EP is a transient wave complex elicited by a cessing in relation to attentional and memory mecha-
certain stimulus or event that is, to be precise, nisms and was first described by Sutton et al. (1965).
repeated only once. The averaged transient EP Evidence that is more recent has shown that the P300
reflects a true response if the relevant brain comprises two subcomponents: (i) the P3a, also referred
mechanisms were in their resting states before to as “novelty P3” and (ii) the P3b, also referred to as
each stimulus, and return to their resting states “classical P300.” The P3a is a positive potential having
before the next stimulus. It is consequently its maximum amplitude over frontal/central electrode
assumed that the EP response to a single event sites and a peak latency in the range of 250–280 ms. This
does not depend on a previous one.” wave has been associated with engaging attention (espe-
cially the orienting, involuntary shifts to changes in the
When the stimuli are, for example, visual, auditory,
environment) and processing novelty. In contrast, the
somatosensory, or even olfactory, then they are called
P3b is a positive potential having the maximum around
evoked potentials. However, ERPs can also be elicited
300 ms, and though dependent on the task, the peak can
by actions that are generated by a person’s internal voli-
vary in latency from 250 to 500 ms. Amplitudes are
tion to perform a task, e.g., when starting a movement, or
typically highest over midline parietal brain areas.
even when such a single movement is attempted or imag-
Generally, the P3b is related to the likelihood of events,
ined (Shibasaki et al., 1980). This particular example of
and the less likely an event, the larger the P3b (Katayama
ERP illustrates the movement-related cortical potential
and Polich, 1998; Simons et al., 2001).
(MRCP). Another example of an ERP is the error-related
The P300-component was used for the design of one
potential (ErrP). After a stimulus, which seems erroneous
of the first BCI systems. Already in 1988, Farwell and
to a person due to a mismatch to the person’s expectation,
Donchin (1988) presented a first BCI paradigm based
the elicited brain wave can be recorded from the midline
on a visual P300. Subjects were presented with a matrix
of the scalp (Falkenstein et al., 1991). The steady-state
of characters displayed on a computer screen. The rows
evoked potential (SSEP) is another type of potential that
and column of the matrix were highlighted by flashes at
gets evoked only if a stimulus (visual or tactile) is
short intervals. Subjects were asked to focus on the
presented with a high repetition rate, usually higher than
desired character (as an intersection of a row and column)
6 Hz. Regan (1989) defines:
and count internally its flashing occurrences (for each
“SSEPs occur when sensory stimuli are repetitively row and column) (see Fig. 18.4A). Since then, many
delivered at high enough rates so that the relevant BCI examples have been and are still published, explor-
neuronal structures are prevented to return to ing the different properties of the P300. One major
their resting states. [..] Ideally, the discrete fre- contribution was made by Kaufmann et al. (2011). Kauf-
quency components remain constant in amplitude mann introduced pictures of faces as stimuli instead of
and phase within an infinitely long time period. simple flashing characters since faces are known for eli-
[..] In practice, SSEPs never completely fulfil this citing particularly strong ERPs due to their psychologic
definition of an ideal SSEP.” salience. In 2013 he published a work where it was
 ELECTROENCEPHALOGRAPHY 257
shown that this system increases the accuracy of ALS polarity, came up. It was described as a characteristic
patients significantly (Kaufmann et al., 2013). wave complex measurable on frontal midline electrodes
Since then, many applications have been developed, above the anterior cingulate cortex (ACC), a brain region
to name some of them: brain painting, web browsing, known for its functional role in conflict monitoring and
music composing (Pinegger et al., 2017). error processing (Carter, 1998; Botvinick et al., 1999,
Since many people are notable for focusing their 2004). ERN develops concurrently with response onset
vision very precisely, tactile stimuli can be used to elicit and often peaks within 100 ms after onset; however,
the P300 (Brouwer and van Erp, 2010; Herweg depending on the type of error peak, latencies may also
et al., 2016). range up until 500 ms after response onset. Generally, the
Moreover, auditory P300 BCIs have also been ErrP occurs after one perceives erroneous events.
developed and now give additional ways to help or assist Depending on the way these potentials are generated
end-users with limited visual pathways (Hohne et al., they are defined as either observation (Miltner et al.,
2010; Schreuder et al., 2011; Pokorny et al., 2013). 1997), feedback (Miltner et al., 1997), response
(Falkenstein et al., 1991), or interaction ErrPs (Ferrez
STEADY-STATE EVOKED POTENTIAL and Millán, 2008; Chavarriaga et al., 2014). Interaction
ErrPs can be detected at the region over the ACC
This potential appears as sinusoids with the same fre-
(Mathalon et al., 2003) and can be measured after a
quency as the stimulation frequency and sometimes with
person witnesses a false execution of an intended com-
higher frequency and as subharmonics (Herrmann, 2001;
mand. From the user’s perspective, an interaction ErrP
M€ uller-Putz et al., 2005a). Since the early 2000s, steady-
occurs whenever a command is misinterpreted by the
state visual evoked potential (SSVEP)-based BCIs, first
used control interface.
introduced by Middendorf et al. (2000), have been
In contrast to the other types of ErrPs, which either do
researched. Since then, many papers have appeared
not require the involvement of the user or do not emerge
investigating higher numbers of classes (Gao et al.,
in self-paced scenarios, the interaction ErrP seems to be
2003), higher harmonics as shown in Fig. 18.4C
the most promising for increasing performance in BCI
(M€ uller-Putz et al., 2005a), and other features like over-
applications for end-users.
laying stimuli to avoid eye movements (Allison et al.,
By using these interaction ErrPs, the performance of
2008). A new application of an SSVEP BCI was shown
BCIs can be improved by detecting specific reactions to
by Pinegger et al. where they used this kind of analysis to
errors that differ from reactions to correct events. False
check, whether a user was focusing on a P300 spelling
actions can be inhibited leading to increased accuracies
matrix. When the user’s attention moves away, the
in BCI-driven systems. Several studies have already
SSVEP elicited by the row-column flashing also dimin-
mentioned the technical capabilities of error correction
ishes and the spelling can be stopped (Pinegger et al.,
for various paradigms (Ferrez and Millán, 2008). In
2014). Since not all end-users keep control of their eye
common, the paradigms used in these experiments are
movements, M€ uller-Putz presented the idea of using
designed to work well for ErrP processing. Because
steady-state somatosensory evoked potentials
of the discrete nature of the feedback, ErrPs can be
(SSSEPs) to create a BCI based on repetitive tactile stim-
detected easily by evaluating time periods after discrete
uli applied to the two index fingers (M€ uller-Putz et al.,
events.
2006). As a basic concept the proof was provided; how-
However, modern BCI applications are no longer
ever, later attempts did not bring this kind of BCI to a
limited to discrete applications where only one discrete
level, where end-users could fully benefit from it
decision can be made at one given point in time. Instead,
(Breitwieser et al., 2016; Pokorny et al., 2016). Neverthe-
continuously controlled applications gain importance as
less, SSVEPs can also be harnessed in healthy BCI users
they offer a more natural implementation of BCI for
when engaging in cooperative tasks (e.g., game playing)
activities of daily living. Relevant examples are a contin-
(Cruz et al., 2017).
uously moving wheelchair for mobility (Galán et al.,
2008) or moving cars in a computer game (Kreilinger
ERROR POTENTIAL
et al., 2016).
One possible way to increase the BCI performance is to A recent study showed that ErrPs can also be recorded
automatically detect errors from recorded brain signals and utilized during one of these continuous applications.
after reactions to particular decisions, thereby, allowing A continuous feedback in the form of a moving artificial
a BCI system to either correct or inhibit erroneous arm was coupled with additional discrete events such as
commands. triggers and ErrPs were successfully found in offline
In the early 1990s, the idea of ErrP, often also referred analysis (Kreilinger et al., 2012; Omedes et al., 2018),
to as error-related negativity (ERN) due to its negative as well as asynchronously (Lopes Dias et al., 2018).
258 €
G.R. MULLER-PUTZ
An alternative and complementary BCI paradigm was Intracortical research on primates (Nicolelis and
presented by Iturrate and colleagues recently (Iturrate Chapin, 2002) and also humans (Hochberg et al.,
et al., 2015). In their approach a robotic arm executed 2006; Collinger et al., 2013) has shown that brain signals
actions that the participants evaluated as erroneous or contain information about hand movement trajectories in
correct, and exploited correlated brain patterns of this a three-dimensional space. Following a similar research
assessment to learn suitable motor behaviors. goal (of decoding movement trajectories) but from
noninvasive EEG activity, some first results (Bradberry
et al., 2010; Ofner and M€uller-Putz, 2012; Antelis
MOVEMENT-RELATED CORTICAL POTENTIAL
et al., 2013; Ofner and M€uller-Putz, 2015) indicated that
Movement-related cortical potentials (MRCPs) were dis- movement trajectory information can be recovered from
covered in 1965 by Kornhuber and Deecke (1965) as the low-frequency EEG amplitudes. Although these
EEG potentials that precede the electromyography studies report promising results, there is still a lot of
(EMG) onset of voluntary action. These potentials have research necessary before such noninvasive BCIs will
been observed over the motor and sensorimotor areas. In be available for end-users. A more detailed review on this
their study, the flexion of the index finger was investi- topic can be found in M€uller-Putz et al. (2016).
gated. Later, MRCPs have been studied in other types
of movement such as, hand movements (Jochumsen CONCLUSIONS
et al., 2015; Ofner et al., 2017; Pereira et al., 2017;
Schwarz et al., 2018), foot movements (Shibasaki In this chapter EEG was discussed in detail, from the neu-
et al., 1981; do Nascimento et al., 2005), and also in other rophysiologic foundations to EEG phenomena, signal
actions such as walking (Jiang et al., 2015; Sburlea et al., recording methods, and respective signal patterns to be
2015b), see Fig. 18.4B. used in BCI research. With its high temporal resolution
In all types of movements, the MRCP emerges before and its easy applicability EEG has played and will con-
the movement onset with a negative slope (the tinue to play a major role in BCI applications for end-
Bereitschaftspotential, BP) and has a similar distribution users with disabilities and in healthy users (Nijholt
for both foot and hand movements, and starts between et al., 2009; Brunner et al., 2015; G€urk€ok et al., 2017).
1.2 and 0.5 s before the movement onset. The maximum In addition, the combination of EEG with other recording
negativity is reached between 0.5 and 0 s before the EMG techniques like near-infrared spectroscopy (Shin et al.,
onset. In the case of foot movement, the MRCP is visible 2017) or functional magnetic resonance imaging (fMRI)
on the midline precentral region and is symmetrically (Zich et al., 2015; Steyrl et al., 2017) have gained impor-
distributed, while for hand movements the MRCP is tance during the last few years. And finally, high density
localized to the contralateral precentral region. It is con- EEG recordings for source localization receive more and
sidered that the maximum negativity of the MRCP is more attention from those with a particular interest in
related to the movement, whereas the BP is related more movement neurophysiology. This is due to the fact that
to a nonspecific preparation (or planning) of the cerebral EEG, in contrast to fMRI, allows measurements while
cortex for voluntary movement (Shibasaki et al., 1980). participants execute body movements (Seeber et al.,
MRCPs are not only observed during movement exe- 2016; Wagner et al., 2016).
cution but are also present during the attempt to execute
and movement imagination. This corresponds to a very REFERENCES
valuable property that can be exploited for BCIs based Allison BZ et al. (2008). Towards an independent brain-
on EEG. MRCPs have been investigated especially in computer interface using steady state visual evoked
the lower delta frequency for the detection of movement potentials. Clin Neurophysiol 119: 399–408.
intention in healthy subjects (Niazi et al., 2011; López- Antelis JM et al. (2013). On the usage of linear regression
Larraz et al., 2014; Jiang et al., 2015; Jochumsen et al., models to reconstruct limb kinematics from low frequency
2015; Sburlea et al., 2015a; Pereira et al., 2017; Pereira EEG signals. PLoS One 8: e61976.
et al., 2018) and in patients (Sburlea et al., 2015a; Bedard C, Kr€ oger H, Destexhe A (2004). Modeling extracel-
Miltner et al., 2016; Sburlea et al., 2017). Ongoing studies lular field potentials and the frequency-filtering properties
of extracellular space. Biophys J 86: 1829–1842.
that focus on the rehabilitation of motor impaired patients €
Berger H (1929). Uber das Elektrenkephalogramm des
(stroke or spinal cord injured) use MRCPs to control Menschen. Arch Psychiatr Nervenkr 87: 527–570.
assistive robotic devices or neuroprosthesis (López- Birbaumer N et al. (1999). A spelling device for the paralysed.
Larraz et al., 2016; M€ uller-Putz et al., 2017). The detec- Nature 398: 297–298.
tion of the MRCP’s earliest component (BP) can also Blankertz B et al. (2008). Optimizing spatial filters for robust
enhance positive neuroplasticity (Mrachacz-Kersting EEG single-trial analysis. IEEE Signal Process Mag 25:
et al., 2012; Xu et al., 2014) and facilitate the recovery. 41–56.
 ELECTROENCEPHALOGRAPHY 259
Blankertz B et al. (2010). Neurophysiological predictor of Daly I et al. (2015). FORCe: fully online and automated arti-
SMR-based BCI performance. Neuroimage 51: fact removal for brain-computer interfacing. IEEE Trans
1303–1309. https://doi.org/10.1016/j.neuroimage.2010. Neural Syst Rehabil Eng 23: 725–736.
03.022. do Nascimento OF, Nielsen KD, Voigt M (2005). Relationship
Botvinick M et al. (1999). Conflict monitoring versus between plantar-flexor torque generation and the magni-
selection-for-action in anterior cingulate cortex. Nature tude of the movement-related potentials. Exp Brain Res
402: 179–181. 160: 154–165.
Botvinick MM, Cohen JD, Carter CS (2004). Conflict moni- Einevoll GT et al. (2013). Modelling and analysis of local field
toring and anterior cingulate cortex: an update. Trends potentials for studying the function of cortical circuits. Nat
Cogn Sci 8: 539–546. Rev Neurosci 14: 770–785.
Bradberry TJ, Gentili RJ, Contreras-Vidal JL (2010). Fabiani M, Gratton G, Federmeier KD (2007). Event-related
Reconstructing three-dimensional hand movements from brain potentials: methods, theory, and applications. In:
noninvasive electroencephalographic signals. J Neurosci J Cacioppo, L Tassinary, G Berntson (Eds.), Handbook of
30: 3432–3437. psychophysiology, Cambridge University Press, pp. 85–119.
Breitwieser C, Pokorny C, M€uller-Putz GR (2016). A hybrid Falkenstein M et al. (1991). Effects of crossmodal divided
three-class brain-computer interface system utilizing attention on late ERP components II. Error processing in
SSSEPs and transient ERPs. J Neural Eng 13: 066015. choice reaction tasks. Electroencephalogr Clin
Brouwer A-M, van Erp JBF (2010). A tactile P300 brain- Neurophysiol 78: 447–455.
computer interface. Front Neurosci 4: https://doi.org/ Faller J et al. (2012). Autocalibration and recurrent adaptation:
10.3389/fnins.2010.00019. towards a plug and play online ERD-BCI. IEEE Trans
Brunner C et al. (2015). BNCI Horizon 2020: towards a road- Neural Syst Rehabil Eng 20: 313–319.
map for the BCI community. Brain Comput Interfaces 2: Farwell LA, Donchin E (1988). Talking off the top of your
1–10. head: toward a mental prosthesis utilizing event-related
Buzsáki G, Draguhn A (2004). Neuronal oscillations in brain potentials. Electroencephalogr Clin Neurophysiol
cortical networks. Science 304: 1926–1929. 70: 510–523.
Buzsáki G, Anastassiou CA, Koch C (2012). The origin of Ferrez PW, Millán JDR (2008). Error-related EEG potentials
extracellular fields and currents—EEG, ECoG, LFP and generated during simulated brain-computer interaction.
spikes. Nat Rev Neurosci 13: 407–420. IEEE Trans Biomed Eng 55: 923–929.
Cabrera AF, Farina D, Dremstrup K (2010). Comparison of Friedrich EVC, Neuper C, Scherer R (2013). Whatever
feature selection and classification methods for a brain– works: a systematic user-centered training protocol to
computer interface driven by non-motor imagery. Med optimize brain-computer interfacing individually. PLoS
Biol Eng Comput 48: 123–132. https://doi.org/10.1007/ one 8: e76214.
s11517-009-0569-2. Galán F et al. (2008). A brain-actuated wheelchair: asynchronous
Cahn BR, Polich J (2006). Meditation states and traits: EEG, and non-invasive brain-computer interfaces for continuous
ERP, and neuroimaging studies. Psychol Bull 132: 180–211. control of robots. Clin Neurophysiol 119: 2159–2169.
Carter CS (1998). Anterior cingulate cortex, error detection, Gao X et al. (2003). A BCI-based environmental controller for
and the online monitoring of performance. Science 280: the motion-disabled. IEEE Trans Neural Syst Rehabil Eng
747–749. 11: 137–140.
Chavarriaga R, Sobolewski A, Millán JDR (2014). Errare Gevins A, Smith ME (2006). Electroencephalography (EEG)
machinale est: the use of error-related potentials in brain- in neuroergonomics. In: R Parasuraman, M Rizo (Eds.),
machine interfaces. Front Neurosci 8: 208. Neuroergonomics. Oxford University Press pp. 15–31.
Cincotti F et al. (2003). The use of EEG modifications due to Goldman MS (2004). Enhancement of information trans-
motor imagery for brain–computer interfaces. IEEE Trans mission efficiency by synaptic failures. Neural Comput
Neural Syst Rehabil Eng 11: 131–133. https://doi.org/ 16: 1137–1162.
10.1109/tnsre.2003.814455. Graimann B et al. (2002). Visualization of significant
Collinger JL et al. (2013). High-performance neuroprosthetic ERD/ERS patterns in multichannel EEG and ECoG data.
control by an individual with tetraplegia. Lancet 381: Clin Neurophysiol 113: 43–47.
557–564. Grimnes S, Martinsen ØG (2000). Data and models. In:
Collura TF (1993). History and evolution of electroencephalo- Bioimpedance and bioelectricity basics, Academic Press
graphic instruments and techniques. J Clin Neurophysiol 195–239.
10: 476–504. Guger C, Ramoser H, Pfurtscheller G (2000). Real-time EEG
Coyle D et al. (2015). Sensorimotor modulation assessment analysis with subject-specific spatial patterns for a brain-
and brain-computer Interface training in disorders of con- computer interface (BCI). IEEE Trans Rehabil Eng 8:
sciousness. Arch Phys Med Rehabil 96: S62–S70. https:// 447–456.
doi.org/10.1016/j.apmr.2014.08.024. Guger C et al. (2012). Comparison of dry and gel based elec-
Cruz I et al. (2017). Kessel run—a cooperative multiplayer trodes for p300 brain-computer interfaces. Front Neurosci
SSVEP BCI game. In: International conference on intelli- 6: 60.
gent technologies for interactive entertainment, Springer, G€
urk€ok H et al. (2017). Meeting the expectations from brain-
Cham, pp. 77–95. computer interfaces. Comput Entertain 15: 5.
260 €
G.R. MULLER-PUTZ
Harmon-Jones E, Beer JS (2012). Methods in social neurosci- Kaufmann T et al. (2011). Flashing characters with famous
ence, Guilford Press. faces improves ERP-based brain-computer interface per-
Harmony T et al. (1996). EEG delta activity: an indicator of formance. J Neural Eng 8: 056016.
attention to internal processing during performance of Kaufmann T et al. (2013). Face stimuli effectively prevent
mental tasks. Int J Psychophysiol 24: 161–171. https:// brain-computer interface inefficiency in patients with
doi.org/10.1016/s0167-8760(96)00053-0. neurodegenerative disease. Clin Neurophysiol 124:
Hashimoto Y, Ushiba J (2013). EEG-based classification of 893–900.
imaginary left and right foot movements using beta Klimesch W (1999). EEG alpha and theta oscillations reflect
rebound. Clin Neurophysiol 124: 2153–2160. https://doi. cognitive and memory performance: a review and analysis.
org/10.1016/j.clinph.2013.05.006. Brain Res Brain Res Rev 29: 169–195.
Herrmann CS (2001). Human EEG responses to 1–100 Hz Kornhuber HH, Deecke L (1965). Hirnpotential€anderungen
flicker: resonance phenomena in visual cortex and their bei Willk€ urbewegungen und passiven Bewegungen des
potential correlation to cognitive phenomena. Exp Brain Menschen: Bereitschaftspotential und reafferente
Res 137: 346–353. Potentiale. Pflugers Arch Gesamte Physiol Menschen
Herweg A et al. (2016). Wheelchair control by elderly Tiere 284: 1–17.
participants in a virtual environment with a brain-computer Kreilinger A, Neuper C, M€ uller-Putz GR (2012). Error poten-
interface (BCI) and tactile stimulation. Biol Psychol 121: tial detection during continuous movement of an artificial
117–124. arm controlled by brain-computer interface. Med Biol Eng
Hobson JA, Pace-Schott EF (2002). The cognitive neurosci- Comput 50: 223–230.
ence of sleep: neuronal systems, consciousness and learn- Kreilinger A et al. (2013). BCI and FES training of a spinal
ing. Nat Rev Neurosci 3: 679–693. cord injured end-user to control a neuroprosthesis.
Hochberg LR et al. (2006). Neuronal ensemble control of Biomed Tech (Berl) 58. https://doi.org/10.1515/bmt-
prosthetic devices by a human with tetraplegia. Nature 2013-4443.
442: 164–171. Kreilinger A, Hiebel H, M€ uller-Putz GR (2016). Single versus
Hodgkin AL, Huxley AF (1952). A quantitative description of multiple events error potential detection in a BCI-
membrane current and its application to conduction and controlled car game with continuous and discrete feedback.
excitation in nerve. J Physiol 117: 500–544. IEEE Trans Biomed Eng 63: 519–529.
Hodgkin AL, Huxley AF (1990). A quantitative description of Leeb R et al. (2007). Self-paced (asynchronous) BCI control of
membrane current and its application to conduction and a wheelchair in virtual environments: a case study with a
excitation in nerve. Bull Math Biol 52: 25–71 (discussion tetraplegic. Comput Intell Neurosci 2007: 79642.
5–23). Linden H et al. (2011). Modeling the spatial reach of the LFP.
Hohne J et al. (2010). Two-dimensional auditory p300 speller Neuron 72: 859–872.
with predictive text system. Conf Proc IEEE Eng Med Biol Logothetis NK, Kayser C, Oeltermann A (2007). In vivo mea-
Soc 2010: 4185–4188. surement of cortical impedance spectrum in monkeys:
Hughes JR (2008). Gamma, fast, and ultrafast waves of the implications for signal propagation. Neuron 55: 809–823.
brain: their relationships with epilepsy and behavior. Lopes Dias C, Sburlea AI, M€ uller-Putz GR (2018). Masked
Epilepsy Behav 13: 25–31. and unmasked error-related potentials during continuous
Iturrate I et al. (2015). Teaching brain-machine interfaces as an control and feedback. J Neural Eng 15: 036031.
alternative paradigm to neuroprosthetics control. Sci Rep 5: López-Larraz E et al. (2014). Continuous decoding of move-
13893. ment intention of upper limb self-initiated analytic move-
Jasper HH (1958). The ten-twenty electrode system of the ments from pre-movement EEG correlates. J Neuroeng
international federation. Electroencephalogr Clin Rehabil 11: 153.
Neurophysiol 10: 371–375. López-Larraz E et al. (2016). Control of an ambulatory
Jeunet C, N’Kaoua B, Lotte F (2016). Advances in user-training exoskeleton with a brain-machine Interface for spinal cord
for mental-imagery-based BCI control. Prog Brain Res 228: injury gait rehabilitation. Front Neurosci 10: 359.
3–35. https://doi.org/10.1016/bs.pbr.2016.04.002. Lorente de No R (1947). Action potential of the motoneurons
Jiang N et al. (2015). A brain-computer interface for single- of the hypoglossus nucleus. J Cell Comp Physiol 29:
trial detection of gait initiation from movement related 207–287.
cortical potentials. Clin Neurophysiol 126: 154–159. Luck SJ (2014). An introduction to the event-related potential
Jochumsen M et al. (2015). Detecting and classifying technique, MIT Press.
movement-related cortical potentials associated with hand Mason SG, Birch GE (2000). A brain-controlled switch for
movements in healthy subjects and stroke patients from asynchronous control applications. IEEE Trans Biomed
single-electrode, single-trial EEG. J Neural Eng 12: 056013. Eng47: 1297–1307. https://doi.org/10.1109/10.871402.
Kaiser V et al. (2014). Cortical effects of user training in a Mathalon DH, Whitfield SL, Ford JM (2003). Anatomy of an
motor imagery based brain–computer interface measured error: ERP and fMRI. Biol Psychol 64: 119–141.
by fNIRS and EEG. Neuroimage 85: 432–444. https:// Melinscak F, Montesano L, Minguez J (2016). Asynchronous
doi.org/10.1016/j.neuroimage.2013.04.097. detection of kinesthetic attention during mobilization of
Katayama J, Polich J (1998). Stimulus context determines P3a lower limbs using EEG measurements. J Neural Eng 13:
and P3b. Psychophysiology 35: 23–33. 016018.
 ELECTROENCEPHALOGRAPHY 261
Melnik A et al. (2017). Systems, subjects, sessions: to what Nicolelis MAL, Chapin JK (2002). Controlling robots with the
extent do these factors influence EEG data? Front Hum mind. Sci Am 287: 46–53.
Neurosci 11: 150. Niedermeyer E (1997). Alpha rhythms as physiological and
Middendorf M et al. (2000). Brain-computer interfaces based abnormal phenomena. Int J Psychophysiol 26: 31–49.
on the steady-state visual-evoked response. IEEE Trans Nijboer F et al. (2015). Usability of three electroencephalo-
Rehabil Eng 8: 211–214. gram headsets for brain–computer interfaces: a within
Millán JDR, Mourino J (2003). Asynchronous BCI and local subject comparison. Interact Comput 27: 500–511.
neural classifiers: an overview of the adaptive brain inter- Nijholt A, Plass-Oude Bos D, Reuderink B (2009). Turning
face project. IEEE Trans Neural Syst Rehabil Eng 11: shortcomings into challenges: brain–computer interfaces
159–161. https://doi.org/10.1109/tnsre.2003.814435. for games. Entertain Comput 1: 85–94.
Millán JDR et al. (2002). A local neural classifier for the rec- Nunez PL, Srinivasan R (2006). Electric fields of the brain: the
ognition of EEG patterns associated to mental tasks. IEEE neurophysics of EEG, Oxford University Press, USA.
Trans Neural Netw 13: 678–686. Obermaier B et al. (2001). Information transfer rate in a
Miltner WH, Braun CH, Coles MG (1997). Event-related brain five-classes brain-computer interface. IEEE Trans Neural
potentials following incorrect feedback in a time- Syst Rehabil Eng 9: 283–288.
estimation task: evidence for a “generic” neural system Ofner P, M€ uller-Putz GR (2012). Decoding of velocities and
for error detection. J Cogn Neurosci 9: 788–798. positions of 3D arm movement from EEG. Conf Proc
Miltner WHR, Bauder H, Taub E (2016). Change in movement- IEEE Eng Med Biol Soc 2012: 6406–6409.
related cortical potentials following constraint-induced Ofner P, M€ uller-Putz GR (2015). Using a noninvasive decoding
movement therapy (CIMT) after stroke. Z Psychol 224: method to classify rhythmic movement imaginations of the
112–124. arm in two planes. IEEE Trans Biomed Eng 62: 972–981.
Mota AR et al. (2013). Development of a quasi-dry electrode for Ofner P, Schwarz A, Pereira J et al. (2017). Upper limb
EEG recording. Sensors Actuators A Phys 199: 310–317. movements can be decoded from the time-domain of
Mrachacz-Kersting N et al. (2012). Precise temporal associa- low-frequency EEG. PLoS One 12: e0182578.
tion between cortical potentials evoked by motor imagina- Omedes J et al. (2018). Factors that affect error potentials
tion and afference induces cortical plasticity. J Physiol 590: during a grasping task: toward a hybrid natural movement
1669–1682. decoding BCI. J Neural Eng 15: 046023.
M€uller-Putz GR, Scherer R, Brauneis C et al. (2005a). Steady- Oostenveld R, Praamstra P (2001). The five percent electrode
state visual evoked potential (SSVEP)-based communica- system for high-resolution EEG and ERP measurements.
tion: impact of harmonic frequency components. J Neural Clin Neurophysiol 112: 713–719.
Eng 2: 123–130. Pereira J et al. (2017). EEG neural correlates of goal-directed
M€uller-Putz GR, Scherer R, Pfurtscheller G et al. (2005b). movement intention. Neuroimage 149: 129–140.
EEG-based neuroprosthesis control: a step towards clinical Pereira J, Sburlea AI, M€ uller-Putz GR (2018). EEG patterns of
practice. Neurosci Lett 382: 169–174. self-paced movement imaginations towards externally-
M€uller-Putz GR et al. (2006). Steady-state somatosensory cued and internally-selected targets. Sci Rep 8: 13394.
evoked potentials: suitable brain signals for brain-computer Pfurtscheller G, Aranibar A (1977). Event-related cortical
interfaces? IEEE Trans Neural Syst Rehabil Eng 14: 30–37. desynchronization detected by power measurements of scalp
M€uller-Putz GR et al. (2007). Event-related beta EEG-changes EEG. Electroencephalogr Clin Neurophysiol 42: 817–826.
during passive and attempted foot movements in paraplegic Pfurtscheller G, Lopes da Silva FH (1999). Event-related
patients. Brain Res 1137: 84–91. EEG/MEG synchronization and desynchronization: basic
M€uller-Putz GR et al. (2016). From classic motor imagery to principles. Clin Neurophysiol 110: 1842–1857.
complex movement intention decoding: the noninvasive Pfurtscheller G, Neuper C (1997). Motor imagery activates pri-
Graz-BCI approach. Prog Brain Res 228: 39–70. mary sensorimotor area in humans. Neurosci Lett 239
M€uller-Putz GR et al. (2017). Towards non-invasive EEG- (2–3): 65–68.
based arm/hand-control in users with spinal cord injury. Pfurtscheller G, Stancák A, Neuper C (1996). Event-related
In: 2017 5th international winter conference on brain- synchronization (ERS) in the alpha band—an electrophys-
computer interface (BCI), IEEE. https://doi.org/10.1109/ iological correlate of cortical idling: a review. Int
iww-bci.2017.7858160. J Psychophysiol 24: 39–46.
Munzert J, Lorey B, Zentgraf K (2009). Cognitive motor Pfurtscheller G et al. (1997). EEG-based discrimination
processes: the role of motor imagery in the study of motor between imagination of right and left hand movement.
representations. Brain Res Rev60: 306–326. https://doi.org/ Electroencephalogr Clin Neurophysiol 103: 642–651.
10.1016/j.brainresrev.2008.12.024. Pfurtscheller G, Neuper C, Krausz G (2000). Functional
Neuper C et al. (2009). Motor imagery and action observation: dissociation of lower and upper frequency mu rhythms in
modulation of sensorimotor brain rhythms during mental relation to voluntary limb movement. Clin Neurophysiol
control of a brain-computer interface. Clin Neurophysiol 111: 1873–1879.
120: 239–247. Pfurtscheller G et al. (2003). ‘Thought’—control of functional
Niazi IK et al. (2011). Detection of movement intention from electrical stimulation to restore hand grasp in a patient with
single-trial movement-related cortical potentials. J Neural tetraplegia. Neurosci Lett 351: 33–66. https://doi.org/
Eng 8: 066009. 10.1016/s0304-3940(03)00947-9.
262 €
G.R. MULLER-PUTZ
Pichiorri F et al. (2011). Sensorimotor rhythm-based brain– Scherer R et al. (2015). Individually adapted imagery
computer interface training: the impact on motor cortical improves brain-computer interface performance in end-
responsiveness. J Neural Eng8: 025020. https://doi.org/ users with disability. PloS One 10: e0123727.
10.1088/1741-2560/8/2/025020. Schl€ogl A et al. (2007). A fully automated correction method
Pinegger A et al. (2014). Write, read and answer emails with a of EOG artifacts in EEG recordings. Clin Neurophysiol
dry “n” wireless brain-computer interface system. Conf 118: 98–104.
Proc IEEE Eng Med Biol Soc 2014: 1286–1289. Schomer DL, da Silva FL (2012). Niedermeyer’s electroen-
Pinegger A et al. (2016). Evaluation of different EEG acquisi- cephalography: basic principles, clinical applications,
tion systems concerning their suitability for building a and related fields, Lippincott Williams & Wilkins.
brain-computer Interface: case studies. Front Neurosci Schreuder M, Rost T, Tangermann M (2011). Listen, you are
10: 441. writing! Speeding up online spelling with a dynamic
Pinegger A et al. (2017). Composing only by thought: novel auditory BCI. Front Neurosci 5: 112.
application of the P300 brain-computer interface. PLoS Schwarz A et al. (2018). Decoding natural reach-and-grasp
One 12: e0181584. actions from human EEG. J Neural Eng 15: 016005.
Pokorny C et al. (2013). The auditory P300-based single- Seeber M, Scherer R, M€ uller-Putz GR (2016). EEG oscilla-
switch brain-computer interface: paradigm transition from tions are modulated in different behavior-related networks
healthy subjects to minimally conscious patients. Artif during rhythmic finger movements. J Neurosci 36:
Intell Med 59: 81–90. 11671–11681.
Pokorny C, Breitwieser C, M€uller-Putz GR (2016). The role of Shibasaki H et al. (1980). Components of the movement-
transient target stimuli in a steady-state somatosensory related cortical potential and their scalp topography.
evoked potential-based brain-computer interface setup. Electroencephalogr Clin Neurophysiol 49: 213–226.
Front Neurosci 10: 152. Shibasaki H et al. (1981). Cortical potentials associated with
Ramoser H, M€uller-Gerking J, Pfurtscheller G (2000). voluntary foot movement in man. Electroencephalogr
Optimal spatial filtering of single trial EEG during Clin Neurophysiol 52: 507–516.
imagined hand movement. IEEE Trans Rehabil Eng 8: Shin J et al. (2017). Evaluation of a compact hybrid brain-
441–446. computer interface system. Biomed Res Int 2017: 6820482.
Regan D (1989). Human brain electrophysiology: evoked Simons RF et al. (2001). On the relationship of P3a and the
potentials and evoked magnetic fields in science and med- novelty-P3. Biol Psychol 56: 207–218.
icine, Elsevier, Amsterdam 414. Steyrl D et al. (2017). Reference layer adaptive filtering
Sburlea AI et al. (2015a). Detecting intention to walk in stroke (RLAF) for EEG artifact reduction in simultaneous EEG-
patients from pre-movement EEG correlates. J Neuroeng fMRI. J Neural Eng 14: 026003.
Rehabil 12: 113. Sutton S et al. (1965). Evoked-potential correlates of stimulus
Sburlea AI, Montesano L, Minguez J (2015b). Continuous uncertainty. Science 150: 1187–1188.
detection of the self-initiated walking pre-movement state Volosyak I et al. (2010). Brain-computer interface using
from EEG correlates without session-to-session water-based electrodes. J Neural Eng 7: 066007.
recalibration. J Neural Eng 12: 036007. Wagner J et al. (2016). Distinct b band oscillatory networks
Sburlea AI, Montesano L, Minguez J (2017). Advantages of subserving motor and cognitive control during gait adapta-
EEG phase patterns for the detection of gait intention in tion. J Neurosci 36: 2212–2226.
healthy and stroke subjects. J Neural Eng 14: 036004. Wolpaw J, Wolpaw EW (2012). Brain–computer interfaces
Scherer R et al. (2004). An asynchronously controlled EEG- principles and practice, Oxford University Press.
based virtual keyboard: improvement of the spelling rate. Xu R et al. (2014). Enhanced low-latency detection of motor
IEEE Trans Biomed Eng 51: 979–984. intention from EEG for closed-loop brain-computer inter-
Scherer R et al. (2007). The self-paced Graz brain-computer face applications. IEEE Trans Biomed Eng 61: 288–296.
interface: methods and applications. Comput Intell Zander TO et al. (2011). A dry EEG-system for scientific
Neurosci 2007: 79826. research and brain-computer interfaces. Front Neurosci 5: 53.
Scherer R et al. (2008). Toward self-paced brain-computer Zich C et al. (2015). Real-time EEG feedback during simulta-
communication: navigation through virtual worlds. IEEE neous EEG-fMRI identifies the cortical signature of motor
Trans Biomed Eng 55: 675–682. imagery. Neuroimage 114: 438–447.