Introduction to Transcranial Current Stimulation

EEG Basic
Principles
clinical-research
applications and
related fields
 2

EEG Basic Principles,

clinical-research
applications and
related fields

Marta Castellano
Neurotech Expert
 3

1
What is an EEG? 4
The electric brain – from neurons to EEG 4
Strengths and limitations of the EEG 6

What to take into account for an EEG test? 7

2
Measuring EEG 7
Artefacts in the EEG 9
First steps in signal processing 9
Brushing over advanced EEG analysis 10
· Connectivity metrics 10
· Source reconstruction 11
· Setting up an EEG experiment – the brain and the brain’s context 12

Clinical and research applications 17

3
EEG biomarkers in brain disorders 17
Cognitive Neuroscience 17
Mobile EEG in everyday applications 18
Consumer Neuroscience and neuromarketing 18
What can we learn from the raw EEG? 19

4
Related fields: real-time exploitation of the EEG 19
Brain Computer Interfaces 20
Neurofeedback 20

5 References 21
 4

1 What is an EEG?
1.1. The electric brain – from neurons to EEG
The human brain is composed of 100 billion neurons (i.e. 100.000.000.000)
- but its complexity does not only come from the sheer number of neurons. To
put it into perspective, an elephant’s brain has about 250 billion neurons, while
chimps and seals both have about 6 billion neurons. Neurons come in
different shapes and sizes – but they do share a similar structure: a cell body
(soma) that extends into axons and dendrites. Dendrites and axons in turn
contain synapses, which are highly specialized junctions connecting two
neurons [2][18].

Neurons make up 10% of our brain tissue and are regarded as the ‘basic
computational unit’ of the brain due to their ability to transmit electrical
signals [2][10]. Neurons, like muscle cells, are electrically excitable cells that
integrate and transmit signals through the generation of action potentials or
spikes [2]. In contrast to other electrically excitable cells, a neuron produces
action potentials (i.e. brief changes in the voltage of the membrane) that are
rather similar in terms of intensity and duration: a 100mV voltage change that
lasts 1-2ms. These action potentials are triggered by electrochemical signals
called neurotransmitters, such as dopamine or serotonin, that control the
release of ions [2]. Depending on the type of ions that are being released,
synapses can be considered excitatory or inhibitory.

The structure of a neuron provides a direction for the propagation of the

action potential. Namely, neurons typically receive signals from other neurons
through synapses within the dendritic tree (post-synapse, the afferent side),
which accumulates within the soma. Once a threshold is reached, the
electrochemical molecules travel along the axons (pre-synapse, the efferent
side) and generate an action potential. This newly generated action potential
in turn, will spread over the neural tissue and mix with the transmembrane
currents of nearby neurons, generating an electric field [1]. In fact, the
exchange of ions across any excitable membrane is what contributes to an
extracellular electric field. And the electric field spreading across brain tissue
is what we can monitor from the scalp with a specific type of technology:
electroencephalography or EEG. But what is the etymology of EEG?
Electroencephalography literally means ‘writing electrical activity of the brain’,
which gives a hint towards the first EEG technology, which was a set of
galvanometers drawn on paper.

Given that neurons are directional, the relative position of nearby cells will
condition how their electric moves is superposed across brain tissue. Thus,
neurons with similar spatial orientation in the brain with synchronized
activation would produce an electrical activity detectable from the surface of
the scalp, as this electric field would be different than the one generated by
un-synchronized or orthogonally oriented neurons.
 5

“The human brain has

100 billion neurons,
each neuron
connected to 10
thousand other
neurons. Sitting on
your shoulders is the
most complicated
object in the known
universe.”
Michio Kaku
 6

When placing sensors on the scalp to monitor electric fields generated by

neurons (i.e. EEG recording), the electric field we record is a smoothed version
of the original electric field, as the signal attenuates and distorts as it passes
through soft and hard tissues (i.e. bone, skin, CSF, muscles, etc…).

All these distortions in the neural signal before it reaches the electrodes
placed on the scalp are grouped into the concept of volume conduction
effects. Because of these distortions and attenuations through tissue, scalp
sensors can only record brain activity generated at several centimeters below
the recording positions. Or from another perspective, at every position where
you put a sensor on the scalp, the recorded brain activity will reflect a
weighted sum of the underlying brain sources that span about 5-10cm²
around that position [1][5] – which correspond to about 30.000 neurons [9].

Measuring the brain activity of 30k neurons is useful as it reflects how a

population of nearby neurons communicate. However, it is not only the local
activity within a brain area that is important for cognition as most of the
cognitive processes arise from how distant areas interact – the connected
brain. Hence, in the section 2, we will shed light on the relevance of brain
connectivity metrics.

1.2. Strengths and limitations of the EEG

There are several reasons why EEG is an exceptional tool to study neural
underpinnings of cognition and disease. However, there are also reasons why
sometimes it may not be the best tool to analyze the brain and cognition. In
fact, whether you use EEG, MEG or other neuroimaging technology to monitor
the brain depends on the study’s question and resources. Let’s analyze
several aspects of EEG technology:

Strengths

• EEG captures brain activity in a millisecond time scale: Behavior occurs at

several timescales. For example, learning to play a new song on the piano
could take days to weeks while the motor action of playing the piano itself
is separated by milliseconds between notes. Yet if your research focuses on
the study of processes that occur in the millisecond to second timescale,
such as studying language, executive function, sensory processing (i.e.
vision) or emotions, both EEG and MEG can capture such responses.
Specifically, they have the ability to monitor the brain activity with sampling
frequencies of 1000Hz, typically.

• EEG is passive and non-invasive: To monitor electroencephalography,

electrical sensors (the electrodes) are placed on the scalp to constantly
monitor ongoing electrical activity from the brain. Thus, by definition, the
technology is, non-invasive and passive.

• EEG is portable and inexpensive: In contrast to other monitoring

technologies, like MEG or MRI, most EEG devices are lightweight and
portable, allowing for real-world brain monitoring (see section 3.4). Costs will
 7

depend on the EEG technology provider, but you can find EEG tech from 100
Euros to more than 25.000 Euros.

Limitations

• EEG captures brain activity of a 5-10cm²: An intrinsic limitation of EEG is

that it allows for the monitoring brain activity of a 5-10cm² brain patch – or
about 30.000 neurons [9]. This poor spatial resolution is a limiting factor for
this neuroimaging technique.

Overall, in comparison to the available techniques to monitor brain activity

(MEG, MRI, PET, etc..), EEG has a clear advantage of being a portable,
inexpensive and non-invasive tool to monitor brain activity with excellent time
resolution. And only recently, this flexibility is becoming more relevant as EEG
manufacturers are shifting EEG devices towards home monitoring.

Figure 1: Comparison of brain monitoring techniques in terms of their spatial and temporal resolution,
where EEG has a clear advantage.

2 What to take into account for an EEG test?

2.1. Measuring EEG
An EEG scan or an EEG test refers to the procedure of monitoring and
reading electrical brain activity with an EEG device. To monitor electrical brain
 8

activity with EEG, typically an array of sensors (the electrodes) are placed on
the surface of the scalp. The number of electrodes ranges between 3 and
256, and its positioning and naming follows international naming systems –
the most common of them being the 10-20 system [11]. In this system,
electrodes are named according to the brain region where they are placed (i.e.
F=frontal, C=central, etc…), and a number that indicates the distance to the
midline - so that distance between adjacent electrodes are either 10 or 20%
from the nasion/inion or between ears. In this naming system, zero (z) would
correspond to all electrodes positioned in the line between the nasion and the
inion, while indexes 7 and 8 would be located at the temporal regions just on
top of the ears. While other electrode nomenclatures exist (i.e. Modified
Combinatorial Nomenclature, 10-5 system for higher resolution recordings, or
the 10-10 system), they all follow a similar naming system that utilizes
anatomical landmarks as measuring points (see Figure 2).

Figure 2: Electrical signals recorded by an EEG at different temporal resolutions and the electrode
positioning system referred as the international 10-20 system, where the distance between electrodes
are either 10 or 20% distance from the midline (nasion to inion) and the coronal plane between mastoids.

The number of electrodes for an EEG recording depends on the type of

experiment. Typical cognitive neuroscience applications make do with 32-64
electrodes (see section 3.3), while clinical applications (where set-up time is a
constraint) might do well with as few as 8 electrodes (see section 3.2). Many
EEG systems use caps or nets that work as a ‘helmet’, where electrodes are
attached; although some of the systems that are used in clinical settings
typically have the electrodes attached to individual wires.

Electrodes are wired to an amplifier, which increases the voltage between

a pair of electrodes: the recording electrode and the reference. Thus, the
signal collected at each of the EEG locations is in fact a voltage difference
between the electrode of interest and a reference signal. The positioning of
 9

the reference electrode is up to the investigator and depends on what

properties of the EEG are paramount for the study (e.g. whether symmetry
needs to be enforced). Some common placements for reference electrodes
include the linked mastoid, earlobe, or the Cz position [21]. And finally, how
long does an EEG take? As you will see in Section 3, the duration of the EEG
recording depends on the application but it can go anywhere from 3 minutes
to several hours.

2.2. Artefacts in the EEG

EEG electrodes monitor electrical potentials, but not all electric potentials
that you monitor will come from the brain. One of the biggest challenges in
EEG analysis is to recognize and eliminate those artefacts – so do not
underestimate the effort you would need for filtering when analyzing your EEG
data. Sources of artefacts can include the following:

Physiological artefacts: these come from the participant and involve

muscle artefacts (EMG), heart beat (ECG or EKG), sweating (i.e. skin
impedance changes) or eye blinks and movements (EOG).

Technical artefacts: these are produced by equipment or the environment,

like cable movement, loss of connectivity with electrodes or the device, or the
50/60 Hz artefact of the electrical devices around you (i.e. fluorescent lights).

Methodologies to detect and remove artefacts from EEG signals are

numerous and depend on the specific type of artefact. However, in general
terms, if you want to trust your EEG test results, make sure to conduct a
careful EEG preparation with the minimum amount of EEG cleaning as
possible. Remember that, despite all your efforts in cleaning EEG signals,
clean EEG data can only appear if you have a good quality EEG recording to
start with (i.e. Garbage IN, Garbage OUT).

2.3. First steps in signal processing

EEG is rhythmic and often studied as an oscillation, which means it is
analyzed by taking into account different frequency ranges at which the
signals oscillate (the frequency-domain analysis) (Figure 2). EEG oscillations
(or commonly called brain waves) are classified according to a set frequency
ranges: delta (0.5-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz) and
gamma (30-100 Hz) (see Table 1).These oscillations are thought to reflect
various aspects of cortical processing: slow oscillations with strong delta
component are present during sleep and are thought to be crucial for memory
formation [13], while awake states are dominated by beta and gamma
oscillatory activity, which have been correlated with perceptual binding,
attention or multi-sensory integration [14][15]. The brain, however, is a
complex non-linear system, and other methodologies that account for this
complexity are also outlined in the literature [12].
 10

Figure 3: Rhythmic activity of the EEG is analyzed according to its frequency.

TABLE 1: The classification of EEG brain waves:

Delta Up to 4 Hz. Highest in Amplitude and the slowest wave. Seen in

waves: adults in slow wave sleep and in babies. Also in patients
suffering from subcortical, diffuse or deep midline lesions or
metabolic encephalopathy hydrocephalus

Theta Between 4 - 7 Hz. Seen in young children, drowsiness or arousal

waves: in older children and adults. Also found during meditative,
relaxed and creative states.

Alpha Between 7 - 14 Hz. Seen mainly in occipital regions when eyes

waves: are closed and during relaxation. This posterior basic rhythm is
slower than 8 Hz in young children. Also seen in coma patients
not responsive to external stimuli.

Sensorimotor Between 8 - 13 Hz. Seen in the motor cortex and reflects the
/mu rhythm: synchronous firing of motor neurons in rest state. Mu
suppression reflects the motor mirror neuron system - when an
action is observed, the pattern extinguishes.

Beta Between 15 - 30 Hz. Linked to motor behavior, active, busy or

waves: anxious thinking and active concentration. Attenuated with
active movements. May be absent or reduced in areas of
cortical damage.

Gamma Between 30 - 100 Hz. Thought to represent binding of different

waves: populations of neurons together into a network for the purpose
of carrying out a certain cognitive or motor function.

2.4. Brushing over advanced EEG analysis

2.4.1. Connectivity metrics
Connectivity metrics describe relationships across signals (i.e. EEG
electrodes) and can refer to the anatomical relationship (anatomical or structural
connectivity), statistical dependencies across those signals (functional
connectivity) or the causal interactions between them (effective connectivity)
 11

[17]. A good review on the different methodologies that can help one understand
how two brain signals depend on each other can be found in [16].

2.4.2. Source reconstruction

Source reconstruction refers to a set of algorithms that, in combination
with high-density EEG, aim to improve on the spatial resolution of EEG. These
methods intuitively aim to reconstruct the underlying electric fields that yield
the observed EEG. In other words, they calculate from a set of observed EEG
data the causal factors (i.e. electric fields) that produced them, which is
known as the inverse problem (see Table 3). The EEG inverse problem is an
ill-posed problem; thus, there is no unique solution. To reconstruct an
approximate solution, we need regularization techniques, like minimum norm
estimates (MNE) or low-resolution electrical activity tomography (LORETA).
These techniques make assumptions about how the electric field is
distributed and conducted through the head. Although different source
reconstruction methods differ on the modeling assumptions, all methods are
known to improve upon the spatial resolution of EEG, reducing the spatial
resolution to 2-4 cm² [6][7][8].

Note, however, that methodologies like Current Source Density (CSD)

estimates (based on the Surface Laplacian (SL) computation) (see Table 2) are
also known to dramatically reduce volume conduction effects and hence
improve EEG spatial resolution – yet, this method does not model the electric
field distribution.

TABLE 2: The Surface Laplacian (SL) computation

In mathematics, the Laplacian is a differential operator (∇∇ or ∇²) that describes the divergence (the
first ∇) of the gradient (the second ∇) of a function f on a point in the space. This means that the
Laplacian of a function f is defined as follows:

Lap(f ) = ∇² f = ∇∇f

It was first proposed in s. XVIII in the study of movement of objects in outer space, but in fact, it’s
used to measure many physical phenomena, like quantum mechanics, wave propagation or electric
potentials. The starting point in EEG applications is the Ohm’s law:

I=V/R

where I refers to the electrical current that passes through conductor material, V is the voltage
measured between two points and R refers to the resistance of this material. Thus, the Ohm’s law
establishes a relationship between the measured voltage by the EEG sensors (V) and the underlying
electric currents (I) of the brain.
 12

TABLE 2: The Surface Laplacian (SL) computation (continuation)

To put these two concepts together, we start by considering that we can describe the surface of the
brain (where the EEG electrodes are placed), by a continuous variable x and y (see figure below) - a
surface that is discretized into a regular-grid that is indexed through the variables i and j. Thus, an
electrode corresponds to a position i, j in the grid.

The fact that we assume the surface of the scalp can be described by the continuous variables x
and y implies that the Laplacian can be computed as follows:
∂² V ∂² V
Laps (V)= +
∂x² ∂y²

And the fact that we assume that those continuous variables x and y can be discretized by the
variables i and j, which are equidistant, implies that the Laplacian can be computed as follows:

V(i-1,j)+V(i+1,j)+V(i,j-1)+V(i,j+1)–4V(i,j)
Laps (V)|(i,j)≈
h²

where h corresponds to the size of the grid. With this, we are able to approximate the voltage from
electrodes around i+1 and j+1 at the central node iand j. Now, this is not the only way of computing
the Surface Laplacian, but several geometric arrangements between points in the grid can be used to
approximate the Laplacian at different electrode locations. Take for instance the different
configurations below:

The one on the left is usually referred to as Hjorth’s approximation and while it works well for
electrodes located at the center of the EEG sensor grid, it doesn’t account for border effects, which is
what the other two SL approximations aim to address. Other SL approximations would allow for a
five-point approximation (for EEG sensor grids that have more electrodes) or account for unevenly-
spaced electrodes. For further detail, see [33]

2.4.3. Setting up an EEG experiment

– the brain and the brain’s context
In setting up your EEG experiment and considering the millisecond
resolution of the EEG (see Table 4), you may want to include monitoring of
other signals in your experiment, such as eye tracking, motor activity, heart
rate, external stimuli like videos or audio, or even other participants’
simultaneous EEG recordings (what is referred to as hyperscanning).
 13

The ability to synchronize these signals to millisecond resolution makes

EEG an ideal technology to study neural processes associated to a very
particular context. In these types of experiments, the EEG can provide
information about the neural response that occurs at the onset of a particular
event, like the appearance of a visual stimulus and the subsequent change in
heart rate. In fact, this type of experiment is well established in the field and is
commonly referred to as event-related potentials (ERPs), which aim to study
brain responses time-locked to the specific sensory, cognitive, or motor events.
It is difficult to view a single EEG response from a stimulus or event as the ERP
amplitude is usually much smaller than the amplitude of background EEG.
Therefore, scientists would apply several trials and average the results together.
In such case the signal-to-noise ratio increases as the square root of numbers
of averaged trials. This procedure allows to filter out the noise or the rest of
uncountable brain processes and leaves the desired waveform for analysis.
This waveform is known as the ERP. Almost any psychological computerized
test with consequently presented trials can be used for ERP recording.
However, in order to record ERPs, the recording set-up must be accurately
synchronized with both the stimuli presentation and EEG recording devices.

TABLE 3: Maxwell’s equations and the inverse problem

What are inverse and forward problems?

Think about baking bread. The forward problem is how to combine known ingredients into the final
bread elaboration. The inverse problem is how to take the final bread and deconstruct it into the
ingredients. It should be apparent with our bread example that the inverse problem is very often far
more complex than the forward problem.

The inverse problem for EEG is that you’re taking a very small number of measurements about a very
complex phenomenon and you’d like to determine what the most likely underlying phenomenon is.
To understand source reconstruction methodologies, it is crucial to understand Maxwell’s equations.
The process of source estimation in EEG is known as the EEG source localization problem. Solving the
EEG source localization problem requires solving the forward and inverse problems. Solving the
forward problem requires estimating the potentials at the electrodes on the scalp, given some source
distribution inside the head. The forward problem is solved repeatedly for different distributions of the
source(s). The EEG inverse problem is solved using the forward solution to estimate the distribution of
source(s) from an EEG recording.
 14

TABLE 3: Maxwell’s equations and the inverse problem (continuation)

For that, we first need to understand how electric fields spread through the brain or, in other words, the
dynamics of the electric field in various media or tissues (i.e. brain, scalp, CSF, etc.) - a behaviour that
was first described by James Maxwell in 1861, in what are called Maxwell’s equations, that are defined
as follows:

∂D
∇×H= + J Ampere’s law
∂t

∂B
∇×E= – Faraday’s law
∂t

∇×D= ρv Gauss’ law

∇∙B=0 Gauss’ law for magnetism

Where ∇ describes the gradient or divergence. The first equation is also known as Ampere’s law and
describes how a magnetic field will spread on a surface. Thus, H refers to the magnetic field strength
(A/m), ∂D/∂t to the displacement of the electric current density (A/m²) and J to the current density
(A/m²).

Faraday’s law specifies that the electric field (E, in V/m) corresponds to the negative time derivative of
the magnetic field (B, in wb/m²). The last two equations, pose constraints on the electric flux density
(D, in C/m²) and the magnetic flux density (B): Gauss’ law constrains how the electric flux gradient
(∇D) can change in relation to the total charge in that volume (ρυ , the charge density in that volume in
C/ m³). Gauss’ law states that the total magnetic flux gradient on a particular volume (∇B) has to add
up to zero.

These equations will also vary depending on the properties of the tissue/media (i.e. whether is
homogeneous, isotropic, etc…), as these properties will change how the tissue enables the currents to
dissipate. To account for these changes, different tissues are described in terms of their permittivity,
permeability and conductivity (s) - which account for tissue properties.
But, is there any way we can reduce the complexity of our attempt in relating the electric field strength
(E, in V/m) and the voltage at the scalp potentials? Note that with the current nomenclature, Ohm’s law
reads as: J = sE

The first simplification, the quasi-static approximation of Maxwell’s equations, relies on the fact that the
frequencies of interest of the brain signals is rather low (typically below 1kHz). With this, we can
reduce Faraday’s law to ∇xE = 0 and overall, segregate the dynamics of electric fields (E) from the
magnetic fields (B), as well as ignoring the propagation delay between the sources of the electric fields
and the scalp. With this, we can also relate the potential (V) to the electric field strength as follows
E = -∇V, so that the new Ohm’s equation reads as J = s∇V.
 15

TABLE 3: Maxwell’s equations and the inverse problem (continuation)

Next, in describing the current density ( J) in a 3 dimensional plane (x,y,z) and assuming that this
current density is spatially bounded to a volume (which is also isotropic) we can further constrain the
quasi-static approximation, which yields to the Poisson equation of the electric field: ∇J = ∇(s∇V). This
is the forward equation for EEG source localization - and basically relates the dynamics of current
density in a 3D space with the dynamics of the voltage given a conductivity’s.

So far we have been able to reduce the problem of switching between neural electric fields that give
rise to the observed scalp potentials to a single equation. Are we done? If we want to approximate the
brain as an isotropic and homogeneous medium, yes. But if we go back to the idea that different
tissues have different conductivities, we need an approach with dealing with these different tissues.
To solve the inverse problem with more than one type of tissue, we need to define a set of boundary
conditions at the interface of different brain regions. The boundary conditions would define how
currents spread across surfaces with different conductivities, and thus, will define the continuity of the
current from one tissue to another.

In approaching source reconstruction by looking at the electric field distribution, it is clear that one of
the key steps is to obtain a 3‑dimensional representation (x,y,z) of the brain: the head modelling step,
as well as information on the conductive properties of the tissue. Modern approaches to head
modelling consider 5‑7 different tissues and advanced numerical solutions to represent the shape of
the head. More information can be obtained in Jatoi, M. A., & Kamel, N. (2017). Brain Source
Localization Using EEG Signal Analysis. CRC Press.

TABLE 4: 6 steps to carry out a successful EEG experiment

1 Understand what you want 1.1. Define your hypothesis: What exactly are you trying
to study and why to learn?

1.2. Conduct a literature review: Are there similar

published works?

1.3. Define your subject population: What is the age range?

Gender? Handedness? Any pathologies you want to include
or exclude?

2 Design your experiment and 2.1. Define your EEG protocol: Do you have defined electrode
prepare your experimental positioning, behavioral tasks associated with your experiment,
campaign population under study, etc...

2.2. Make sure you have fleshed out the details of your
experimental campaign: Do you have a clear idea of the
possible covariates? Are your information sheets, informed
consent forms and other documents ready?

2.3. Ensure you have all the equipment necessary for the
whole experimental campaign: EEG device, electrodes, gel,
recording computer and stimulus computer if necessary.

2.4. Prepare your lab: make sure you have a calm and silent
environment, ideally only with the necessary equipment (if
possible, computers should be placed outside of the recording
room to avoid any interference). Make sure the recording room
is as comfortable as possible for the participants
 16

3 Test your experimental 3.1. Get familiarized with the experimental protocol and the
protocol before the real recording room: Can you execute the different steps of your EEG
experiment recording without difficulty?

3.2. Foresee all the possible problems that might occur in the
recording session and have backup plans for each of them:
Do you know where all the consumables are placed?
Do you have cleaning materials to easily record several
participants in a row?

3.3. Make sure that all your external sources of signals are
connected and synchronized to your EEG recording (i.e. GSR,
eyetracking, etc…)

3.4. Execute the complete EEG experimental campaign from a

small sample of volunteers (most of the time this is the FFF
step: Fools, Friends and Family). This is the pilot campaign,
which will test all the aspects of your experimental design.

4 Perform a preliminary data 4.1. Check for the EEG data quality: Is the data format correct?
analysis with the data Has the data been properly recorded? Do you have triggers from
collected in the pilot external sources? Do you need accelerometer data? What is the
campaign expected size of your data files?

4.2. Check for the coherence of your experimental campaign:

Is the experimental protocol addressing your question?
Do you need to add more questionnaires to control for
possible confounds?

4.3. Check the participant: Is the recording room comfortable?

Do you have sources of distraction during the recording?

5 Get ready for the 5.1. Participant involvement before starting the recording: Is
experimental campaig your informed consent ready? Do you have a timetable of the
visits of the different participants? Do they have all the
information they need to conduct the experiment?

5.2. Participant involvement at the day of the recording: Note

all of the relevant information that allows you to identify the
participant (i.e. contact info, date and all of the details that can
modify your EEG recording (i.e. electrode impedances from
hair thickness)

5.3. Monitoring experiment covariates: Make sure you have

stored information about the covariates that you identified to
be relevant for your experiment, including age, gender,
handedness, etc...

5.4. Monitoring the EEG recording: Supervise the recording

at all times and write down any details of the recording, such
as movements, falling asleep, scratching, sneezing,
coughing, etc...

5.5. Finalize your experiment: Make sure your equipment is

clean and ready for the next recording.

6 Enjoy the recording - may you gather great data!

 17

3 Clinical and research applications

3.1. EEG biomarkers in brain disorders
So far, we have briefly described some metrics and features that can be
used to analyze our EEG data (i.e. frequency-based metrics, connectivity
metrics). But what if instead of monitoring a particular EEG characteristic,
we directly monitor the trait we are interested in? Take for instance an EEG
recording from all the people that live in your building – wouldn’t you prefer
to know whether anyone will develop Parkinson’s Disease (PD) rather than
knowing the oscillatory power in alpha band of their prefrontal cortex? At
that point, whatever combination of EEG metrics that you have used to
predict the onset of PD is irrelevant - what matters is the clinical trait. The
combination of EEG features becomes a biomarker, which provides
quantitative information about a trait that is useful for many clinical and
research applications.

Current diagnoses of most neurological disorders rely on tests and

questionnaires that characterize the phenotype of the disease, but often not
its cause. Hence, diagnosis for complex neurological disorders, like
Alzheimer’s disease, remains 60-70% accurate. As a result, intensive efforts
in finding quantifiable biomarkers that increase the accuracy of the
diagnosis are being done (i.e. genetic causes, blood tests). But there are
four factors that make EEG an increasingly effective approach to identifying
biomarkers of neurological disorders: increasing accessibility of EEG
devices for home use, wireless capabilities, cost-efficiency, and ongoing
technical advancements that improve its sophistication.

This is particularly useful for complex neurodevelopmental diseases in,

for example, accurate diagnosis of children with Attention Deficit
Hyperactivity Disorder (ADHD) whose clinical criteria diagnoses 5-7% of
children in the USA and 1-2% in the EU, with an overall diagnostic accuracy
of 61% [19]. For that, the theta/beta ratio at frontal electrodes is monitored
in rest-EEG, a quantitative biomarker that is currently under review by the
FDA [20].

3.2. Cognitive Neuroscience

What is cognitive neuroscience? The field of cognitive neurosciences
deals with understanding the mechanisms through which the brain
implements cognition and behavior – the mental process of acquiring
knowledge and understanding through thought, experience, and senses.

Cognitive processes are quick – after all, if we want to survive, we need to

be able to generate a response to the environment in tens to hundreds of
milliseconds – faster than what it takes for a car to crash into a bicyclist. But,
how do we study cognition? Cognitive processes that ultimately lead to
behavior are extensive, and usually a mixture of them are required to execute
a particular task.
 18

Take for example, the act of grasping an object – a spoon from a drawer.
Such a basic motor action involves cognitive processes involved in planning,
preparation, and execution of a motor action, which in turn involve processes
like working memory, attention, or decision-making. All within the few
seconds that it takes you to end that action. Brain oscillatory activity (within
and between diverse cortical areas) has been associated with the execution
of such actions and altered in disease [22]. Other aspects of cognition that
are being largely studied with EEG involve memory [13], attention [14],
multi‑sensory integration [15] or language processing [23].

3.3. Mobile EEG in everyday applications

Over the last decades, EEG devices in the market have become more
sophisticated in their design to optimize usability in out-of-the-lab settings,
starting by the availability of wireless systems. Taking advantage of its portability,
several applications that require free-moving brain monitoring have appeared:
sports monitoring, driving or flying simulators, social gathering (e.g. through
hyper-scanning), consumer neuroscience, marketing research or media testing.

Most of these applications appear not only due to the availability of mobile-
EEG systems, but also from our ability to extract physiological markers from
ongoing EEG. For example, sports monitoring benefits from understanding brain
markers of fatigue or engagement, while marketing applications (consumer
neuroscience, marketing research or media testing) classically study emotional
states in terms of arousal or valence. Some mobile-EEG technology already
includes the extraction and visualization of these cognitive-emotional metrics.

3.4. Consumer Neuroscience and neuromarketing

Consumer neuroscience refers to the combination of neuroscience with
marketing research where the process of linking consumer needs to producers is
driven by the neuroscientific knowledge of consumer behavior. The study of
consumer neuroscience is broad and classically involves a combination of
psychologists, sociologists and anthropologists. Neuropsychology contributes to
this aim through the study of decision-making processes, attention or affective
responses [28]. When this research is focused on the understanding of marketing
techniques (i.e. packaging, websites or branding), it is referred to as
neuromarketing.

Neuromarketing aims to find effective and efficient marketing strategies

based on the neural response of consumers. A typical neuromarketing
experiment would use brain-monitoring techniques (such as EEG) to complement
traditional methods of marketing by providing emotional and behavioral metrics.

Additionally, the ability to extract cognitive states like engagement,

drowsiness, or motivation from EEG recordings can be useful in analyzing
consumer behavior. For instance, EEG has been used to monitor emotional
responses to paintings exhibited in a museum [29], or the arousal levels of
different perfumes [30].
 19

3.5. What can we learn from the raw EEG?

Is there anything we can learn by directly observing an EEG recording? Is
there any direct interpretation that can be informative to us?

In sleep labs, the raw EEG data is examined to assess the different stages
of sleep. The single most important laboratory technique for assessment of
sleep and its disorders is called polysomnography (PSG). PSG consists of
recordings of multiple physiological characteristics during sleep in tandem with
EEG, including eye movements, muscle activity, multiple breathing variables and
blood oxygen levels during sleep. For the study of sleep patterns, the EEG is
examined on a scale of 30 seconds to minutes while sleep-specific EEG
features are studied. In particular, sleep researchers focus on the properties of
different oscillations: delta bands (1-4Hz) – the slow waves generated during
sleep – appear during non-REM together with spindles (bursts of oscillations at
12-14Hz and a duration of 0.5 to 1.5 seconds), while REM sleep is characterized
by a prominent theta-rhythm. These EEG features (and others) are altered in
various neurological disorders like PD, Fibromyalgia or Dementia. These
alterations accompany changes in sleep patterns. Monitoring with EEG during
sleep can therefore provide insights into the effects associated with sleep-
altering disorders.

In clinical settings, the most common use of EEG is to monitor and diagnose
epileptic activity. Epileptologists, like sleep researchers, monitor EEG activity for
long periods of time (often several hours) but in this case, the main purpose of
the EEG is to assess interictal and ictal epileptiform activity and focus on EEG
monitoring that can help locate the source of the epileptic seizure. EEG is also
used to detect abnormal EEG activity associated with other brain disorders like
head injuries, encephalitis (inflammation of the brain), brain tumors, strokes and
dementia. EEG monitoring is also very important in Intensive Care Units (ICU) and
Emergency Rooms (ER) for quickly differentiating the etiology and therapeutic
efficacy in critically ill patients with a variety of cerebral injuries and altered states
of consciousness. It is also the only procedure to diagnose nonconvulsive status
epilepticus. Furthermore, the determination and diagnostic validation of brain
death requires an EEG recording.

4 Related fields:
real-time exploitation of the EEG
The real-time monitoring of EEG (or features of the EEG) allows for the
design of closed-loop systems, or experiments that utilize the analysis of EEG
signals in real-time.

Drawing from communication theory, in closed-loop systems, the brain

would be the source of information or speaker and the link between the ‘EEG
device and EEG analysis algorithm’ would be the channel that is being used to
transmit the signal/information. Then, to complete the closed-loop, we just
 20

need an adaptive system that respond automatically and immediately to

these brain-states.

Thus, with this point of view in mind, applications for real-time EEG
analysis is vast —but they all depend on the ‘system’ that listens to and
utilizes the brain signal. Below, we present three different concepts— and let
me propose an activity for today: can you turn brain reading in a game?

4.1. Brain Computer Interfaces

Brain Computer Interfaces (BCIs) refers to the direct communication
between brain and an external device. The list of what is considered an external
devices suitable for BCI is interminable, but the field of BCI has mostly focused
on the development of neuro-prosthetic devices that aim to restore or support a
particular function, like walking, seeing or hearing. Following years of
experimentation, laboratories and hospitals working on BCI (a field that gained
relevance in the 1950s) have developed several successful applications, like the
control of a prosthetic robotic limb with 7 degrees of freedom [25] or the
synthesis of speech from cortical activity [26].

Non-invasive brain monitoring like EEG also has success stories, despite its
low spatial resolution. For instance, in the late 1990s, Niels Birbaumer
presented ten amyotrophic lateral sclerosis patients (completely locked-in) who
could generate a binary signal to control a computer through the self-regulation
of their slow cortical potentials [24]. From this binary classifier, the most recent
research reports the decoding of visual information for up to 500k different
visual stimuli [27].

4.2. Neurofeedback
In neurofeedback applications, brain activity is strengthened using video or
sounds with the goal of teaching the viewer to self-regulate that brain function.
As such, brain activity can be monitored by any of the techniques presented in
Figure 2 – and when EEG is used, visualizations display either the raw EEG or
some transformation of it (like the alpha oscillatory power or motivation).

More than 70 clinical studies are being conducted right now to assess the
efficacy and safety of neurofeedback with more than 100 clinical studies
already completed. Typical applications include ADHD, Mild Cognitive
Impairment and epilepsy, although it has been successfully used for the
treatment of pain, autism and addiction.

Typical neurofeedback treatment starts by the identification of the

EEG‑feature of interest (i.e. beta enhancement and theta suppression in ADHD)
and by establishing a way in which this feature will be visualized (usually
through the use of game-like applications). From here onwards, the treatment
consists of allowing the patient to identify the association between brain
function and visual stimulus – and to develop strategies on how to improve it.
 21

Neurofeedback, however, refers to the self-regulation of brain activity and

thus, it has been used to enhance cognitive performance [31] and as an
experimental method to investigate the causal role of specific EEG-features in
behavior – an approach that is also known as brain-state dependent
stimulation [32].

5 References
[1] Buzsáki, G., Anastassiou, C. a., and Koch, C. The origin of extracellular fields and
currents–EEG, ECoG, LFP and spikes. Nature reviews Neuroscience 13, 6 (June
2012), 407–20.
[2] Kandel, E. R., Schwartz, J. H., and Jessell, T. M. Principles of Neural Science, 54th
editi ed. McGraw-Hill Professional, 2012.
[3] Luck (2014, 2nd edition). An introduction to the event-related potential technique.
Cambridge, MA: MIT Press.
[4] Niedermeyer & Lopes da Silva (2012, 6th edition). Electroencephalography: Basic
Principles, Clinical Applications, and Related Fields. Philadelphia, PA: Lippincott
Williams & Wilkins.
[5] Makeig, S., Bell, A., Jung, T.-P., Sejnowski, T., 1996. Independent component analysis
of elec- troencephalographic data. In: Touretzky, D., Mozer, M., Hasselmo, M. (Eds.),
Advances in Neural Information Processing Systems vol. 8. MIT P, Cambridge MA,
pp. 145–151.
[6] Pizzagalli, D. A. (2007). Electroencephalography and high-density
electrophysiological source localization. Handbook of psychophysiology, 3, 56-84.
[7] Yao, J., & Dewald, J. P. (2005). Evaluation of different cortical source localization
methods using simulated and experimental EEG data. Neuroimage,25(2), 369-382.
[8] Ding, L., Lai, Y., & He, B. (2005). Low resolution brain electromagnetic tomography in
a realistic geometry head model: a simulation study. Physics in Medicine and
Biology, 50(1), 45.
[9] Young, N. A., Collins, C. E., & Kaas, J. H. (2013). Cell and neuron densities in the
primary motor cortex of primates. Frontiers in neural circuits, 7, 30.
[10] Ikeda, S., & Manton, J. H. (2009). Capacity of a single spiking neuron channel. Neural
Computation, 21(6), 1714-1748.
[11] American Electroencephalographic Society. (1994). Guideline thirteen: Guidelines for
standard electrode position nomenclature. Journal of Clinical Neurophysiology, 11(1),
111-3.
[12] Rodriguez-Bermudez, G., & Garcia-Laencina, P. J. (2015). Analysis of EEG signals
using nonlinear dynamics and chaos: a review. Applied mathematics & information
sciences, 9(5), 2309.
[13] McCormick, D. a., and Bal, T. Sleep and arousal: thalamocortical mecha- nisms.
Annual review of neuroscience 20 (Jan. 1997), 185–215
[14] Fries, P. The model- and the data-gamma. Neuron 64, 5 (Dec. 2009), 601–2.
[15] Engel, A. K., and Fries, P. (2010). Beta-band oscillations-signalling the status quo?
Curr. Opin. Neurobiol. 20, 156–165. doi: 10.1016/j.conb.2010.02.015.
 22

[16] Dauwels, Justin, et al. “A comparative study of synchrony measures for the early
diagnosis of Alzheimer’s disease based on EEG.” NeuroImage 49.1 (2010): 668-693.
[17] Sporns, Olaf (2007), Brain Connectivity.. Scholarpedia, 2(10):4695.
[18] Niedermeyer & Lopes da Silva (2012, 6th edition). Electroencephalography: Basic
Principles, Clinical Applications, and Related Fields. Philadelphia, PA: Lippincott
Williams & Wilkins.
[19] Snyder SM, Rugino TA, Hornig M, Stein MA. Integration of an EEG biomarker with a
clinician’s ADHD evaluation. Brain Behav. 2015;5: 1–17. doi:10.1002/brb3.330.
[20] Ogrim G, Kropotov J, Hestad K. The quantitative EEG theta/beta ratio in attention
deficit/hyperactivity disorder and normal controls: Sensitivity, specificity, and
behavioral correlates. Psychiatry Res. Elsevier Ltd; 2012;198: 482–488.
[21] Michal Teplan et al. Fundamentals of eeg measurement. Measurement science
review, 2(2):1–11, 2002.
[22] Oliveira, A. S., Arguissain, F. G., & Andersen, O. K. (2018). Cognitive processing for
step precision increases beta and gamma band modulation during overground
walking. Brain topography, 31(4), 661-671.
[23] Castellano, M., Kroupi, K., Acedo J., Borrega, O., Rojas A., de la Torre, C.A., Ruffini, R.,
Soria-Frisch, R., Hinzen W. (in prep) Neural signatures of sentence comprehension
-strategic resource allocation via gamma oscillations.
[24] Birbaumer, N., Ghanayim, N., Hinterberger, T., Iversen, I., Kotchoubey, B., Kübler, A., ...&
Flor, H. (1999). A spelling device for the paralysed. Nature, 398(6725), 297.
[25] Collinger JL, Wodlinger B, Downey JE, et al. High-performance neuroprosthetic control
by an individual with tetraplegia. Lancet. 2013;381(9866):557–564. doi:10.1016/
S0140-6736(12)61816-9
[26] Anumanchipalli, G. K., Chartier, J., & Chang, E. F. (2019). Speech synthesis from neural
decoding of spoken sentences. Nature, 568(7753), 493.
[27] Nagel, S., & Spüler, M. (2019). World’s Fastest Brain-Computer Interface: Combining
EEG2Code with Deep Learning. bioRxiv, 546986.
[28] Lee, N., Broderick, A. J., & Chamberlain, L. (2007). What is ‘neuromarketing’? A
discussion and agenda for future research. International journal ofpsychophysiology,
63(2), 199-204.
[29] J. Acedo, A. Soria-Frisch, D. Ibáñez, M. Castellano, S. Dunne (2015) Affective BCI for
characterizing museum visitors response. Asilomar Conference on Signals, Systems,
and Computers Technically co-sponsored by IEEE Signal Processing Society.
[30] Kroupi, E, (2019) ExperienceLab doesn’t lie: objective access to the inner emotional
world of your user groups in real time. Neuroelectrics blog post.
[31] Gruzelier, J. H. (2014). EEG-neurofeedback for optimising performance. I: a review of
cognitive and affective outcome in healthy participants. Neuroscience & Biobehavioral
Reviews, 44, 124-141.
[32] Enriquez-Geppert, S., Huster, R. J., & Herrmann, C. S. (2017). EEG-neurofeedback as a
tool to modulate cognition and behavior: a review tutorial. Frontiers in human
neuroscience, 11, 51.
[33] Carvalhaes, C., & de Barros, J. A. (2015). The surface Laplacian technique in EEG:
Theory and methods. International Journal of Psychophysiology, 97(3), 174-188.
23