Brain Computer Interface
A Seminar Report
by
Shiju S.S
105115
Department of Computer Science & Engineering
College of Engineering Trivandrum
Kerala - 695016
2010-11
Abstract
A brain computer interface presents a direct communication channel from the brain to the
computer. The BCI processes the brain activity and translates it into system commands using
feature extraction and classi cation algorithms. EEG-based BCI experiments have been de-
signed and conducted. The experiments are designed to nd distinctive brain patterns which
are generated voluntary. Various researches have been going on in EEG Based BCI. While
most current brain computer interface research (BCI) is designed for direct use with disabled
users, This seminar is to explain functional near-infrared spectroscopy (fNIRS), a non- invasive
brain measurement device, to augment an interface so it uses brain activity measures as an
additional input channel. Future work in BCI will focus on creating an interface that responds
to one of those measures by adapting the interface. By combining brain signal measured with
an adaptive interface it is expect to contribute a functional passive brain-computer interface
that measures and adapts to the users brain signal.
[attachment=8559]
1 Introduction
What is a Brain Computer Interface? As mentioned in the preface a BCI represents a direct
interface between the brain and a computer or any other system. BCI is a broad concept
and comprehends any communication between the brain and a machine in both directions:
e ectively opening a completely new communication channel without the use of any peripheral
nervous system or muscles.
In principle this communication is thought to be two way. But present day BCI is mainly
focusing on communication from the brain to the computer. To communicate in the other
direction, inputting information in to the brain, more thorough knowledge is required concerning
the functioning of the brain. Certain systems like implantable hearing-devices that convert
sound waves to electrical signal which in turn directly stimulate the hearing organ already exist
today. These are the rst steps. The brain on the other hand is on a whole other complexity
level compared to the workings of the inner ear.
Figure 1: Basic BCI layout
From here on the focus is on communication directly from the brain to the computer. Most
commonly the electrical activity ( elds) generated by the neurons is measured, this measuring
technique is known as EEG (Electroencephalography). An EEG-based BCI system measures
speci c features of the EEG-activity and uses these as control signals.
Over the past 15 years the eld of BCI has seen a rapidly increasing development rate and
obtained the interest of many research groups all over the world. Currently in BCI-research
the main focus is on people with severe motor disabilities. This target group has little (other)
means of communication and would be greatly assisted by a system that would allow control
by merely thinking.
The concept of thinking is perhaps too broad a concept and can actually better be replaced
by generating brain patterns. The general picture of a BCI thus becomes that the subject
is actively involved with a task which can be measured and recognized by the BCI. This
task consists of the following: evoked attention, spontaneous mental performance or mental
imagination. The BCI then converts the 'command' into input control for a device (see gure
1).
5
This is the basic idea. With the continuously increasing knowledge of the brain and advances
in BCI over time, perhaps BCI will be able to extract actual intentions and thoughts. This
however does not appear to be on the cards for the very near future.
6
2 Concepts
The de nition of BCI as quoted from the rst international meeting devoted to BCI research
in 1999.'A brain-computer interface is a communication system that does not depend on the
brain's normal output pathways of peripheral nerves and muscles'.The goal is to acquire knowl-
edge of the intentions of the user either consciously or unconsciously by means of measurement
of brain activity. This goal can be achieved in various ways, but it all starts with the brain and
thus with the most basic element of the brain.
2.1 The neuron
A neuron is a cell that uses biochemical reactions to receive, process and transmit information.
It consists of the cell body (Soma) in which the cell core (Nucleus) resides (see gure 2). Each
neuron has one axon; this is a long 'cable'-like part of the neuron which is used to reach other
neurons. The soma of a neuron is branched out into dendrites to which axon-ends from other
neurons connect.
Figure 2: Overview of the neuron
The dendrites are not in actual physical contact with the axons of other neurons; a small cleft
exists between them: the synaptic gap. This is the location where the impulse is transferred.
When a neuron res, it sends signals to all the neurons that are connected to its axon via
the dendrites. The dendrites can be connected to thousands of axons; all incoming signals
combined are added through spatial and temporal summation. If the aggregate input reaches
a certain threshold, the neuron will re and send a signal along its own axon. The strength of
this output signal is always the same, no matter the magnitude of the input.
This single signal of a neuron is very weak. The numerous neurons in the brain are constantly
active. The generated activity can be measured. It appears to be impossible to measure the
individual activity of every neuron. Moreover it is questionable whether it would be a real gain,
since neurons work in groups to achieve a certain goal. The activity from a group of neurons
however can be measured. For the signals of neurons to be visible using EEG in particular, a
couple of conditions need to be met, which are summarized schematically in gure 3.
The electrical activity of the neuron must be perpendicular to the scalp in order for the
EEG to fully pick up the eld.
A large number of neurons must re parallel to each other.
7
Figure 3: Cross-cut of the head: only the green neuronal activity can be measured using EEG
The neurons must re in synchrony with the same polarity, in order not to cancel each
other out.
2.2 The Brain
Combining about 100 billion neurons results in what is called the human brain. The brain
consists of the following elements ( gure4)
The brainstem is an important relay station. It controls the re
exes and automatic
functions, like heart rate and blood pressure and also sleep control.
The Cerebellum integrates information about position and movement from the vestibular
system to coordinate limb movement and maintaining equilibrium.
Mid-brain: amongst others the Hypothalamus and pituitary gland control visceral func-
tions, body temperature and behavioral functions like, the body's appetite, sleep patterns,
the sexual drive and response to anxiety, aggression and pleasure.
The Cerebrum (or cerebral cortex) receives and integrates information from all of the
sense organs and controls the motor functions. Furthermore it contains the higher cerebral
functions like: language, cognitive functions and memories. Emotions are also processed
in the cerebrum.
The cortex of the cerebrum is part of the brain which is of the most interest for BCI. It is
responsible for the higher order cognitive tasks and is near the surface of the scalp. In addition
that functionality in the brain appears to be highly local.
The cerebrum is divided into two hemispheres, left and right. The left halve senses and
controls the right half of the body and vice versa. Each hemisphere can be divided into four
lobes, the frontal, the parietal, the occipital and the temporal (see gure4). The cortex can also
by divided in certain areas each of which is specialized for a di erent function. Especially the
sensorimotor cortex is important for BCI. Over this part the entire human body is represented.
8
Figure 4: Brain overview
Figure 5: Homunculus
The size of area corresponds with the importance and complexity of movement of that particular
body part (see gure5).
9
3 Electroencephalogram
The best brain measurement method would have a high spatial and temporal resolution, be
very cheap, portable and easy to apply non-invasively. This method does not (yet) exist.
Of all methods listed in the previous section, EEG is by far the most commonly used in BCI.
The prime reason for this is the excellent temporal resolution which is a necessity for real-time
BCI. And although the spatial data resulting from EEG is often distorted and far from perfect,
EEG o ers direct functional correlation of brain activity.
Another major plus is the ease of applying this method. With a cap containing only a few
electrodes measurements can start. For practical uses and applications it is small and relatively
portable, which improves prospects of future applications.
Aside from the ease of appliance, this is also a relatively low-cost method, certainly compared
to methods like MEG, which require expensive equipment and skilled professionals to operate.
Although EEG is the most commonly used, this does not mean that others methods are
not feasible. With the continuous improvement of the techniques involved, they can become a
viable option in the future.
EEG comes in two
avors; the most commonly used in BCI is the non-invasive variant. The
electrode is placed on the scalp. The obvious advantage is that it can be safely applied to
anyone at any moment without a lot of preparation.
The second variant is the invasive EEG. Instead of attaching the electrode on the skull, it
is placed inside. The advantage of this variant is the higher spatial resolution obtained by it.
With non-invasive EEG, the skull causes signi cant spatial smearing of the measured activity:
leading to more dicult localization of the original signal, which degrades the quality of the
signal.
3.1 10-20 system
A cap with a number of electrodes is placed on the user's head. At the TU Delft the 10-20
system of electrode placement is used. This is an international standard used for comparing
results among di erent research. The system is based on the relationship of the electrode
placement and the underlying area of the cerebral cortex. Each location on the scalp has a
letter to identify the hemisphere location (Frontal, Temporal, Central, Parietal and Occipital
Lobe) and a number to de ne the hemisphere. Ranging from 1 to 8, with the even number
referring to the right hemisphere and the odd numbers to the left hemisphere (see gure 6).
The 10-20 refers to the distance (in percentage) between the di erent electrodes. Reference is
needed to measure voltage. Reference electrodes are usually attached to relative stable points
where the potential remains constant. Points like the earlobes or mastoid bones behind the
ear.
This a-periodic and unpredictable activity is constantly present and is a result of the total
activity generated by all the neurons in the brain. The frequency range is divided into di erent
band: The Delta (0.1-3-5Hz), Theta (4-7.5Hz), Alpha (8-13Hz), Beta (14-30Hz) and Gamma
10
Figure 6: The international 10-20 system
(>30Hz)(see gure 7). The mu-rhythm is a speci c part of the Alpha rhythm (10-12Hz) and is
located over the sensorimotor cortex. The main advantage of the mu-rhythm over the Alpha
rhythm is that is does not appear to be in
uenced by eye-blinking therefore it is mainly used
in BCI. Users can learn to voluntary control the rhythms after training to some extent. This
concerns the synchronization of the rhythm.
Figure 7: Overview of the categorization of brain waves
Motor Imagery is a commonly used method in BCI. To obtain MI, the user should imagine
moving a hand, nger or leg but not actually moving it. Thereby generating the pattern in the
brain that goes with this movement, but not disturbing the EEG measurement by the actual
activity of muscles. Measurement of muscle activity is called EMG (electromyography) and
this activity overwhelms the EEG.
3.2 Artifacts
The EEG signals are always imperfect and always contaminated with artifacts. Artifacts are
undesirable disturbances in the signal. These artifacts range from bioelectrical potentials pro-
duced by movement of body parts like, eyes, tongue, arms or heart or
uctuation in skin re-
sistance(sweating). And can also have a source out side the body like interference of electrical
equipment nearby or varying impedance of the electrodes.
11
3.3 Artifact removal
Whenever artifacts are detected the a ected portion of the signal can be rejected. This can
be a valid pre-processing step and does not have to be a problem. However the problem with
deleting a speci c piece of data is that it can result in strange anomalies where the two pieces
are connected. Secondly, EEG data in general is relatively scarce. For that reason a better
approach is to remove the artifact from the EEG data. This goes one step further than artifact
rejection.
3.4 Independent Component Analysis
Higher-order statistical methods simultaneously use the information of all the electrodes avail-
able. This o ers the possibility to locate a certain component and remove it from the data. One
method often applied is Independent Component Analysis (ICA) also known as blind source
separation.
ICA is a statistical computational spatial ltering method that decomposes the multi-electrode
data into underlying independent components (or as independent as possible). The goal is to
reveal hidden factors which underlie a certain dataset. ICA assumes linear independence of the
sources and that the sources are a linear combination of the witnessed output. ICA does not
take into account any 'ground-truth'-labels, which makes it an unsupervised method.
12
4 Magnetoencephalogram
Magneto-encephalography (MEG) is a technique for mapping brain activity by recording mag-
netic elds produced by electrical currents occurring naturally in the brain, using arrays of
SQUIDs (superconducting quantum interference devices). Applications of MEG include local-
izing regions a ected by pathology before surgical removal, determining the function of various
parts of the brain, and neurofeedback.
4.1 The basis of the MEG signal
Synchronized neuronal currents induce weak magnetic elds. At 10 femtotesla (fT) for cortical
activity and 103 fT for the human alpha rhythm, the brain's magnetic eld is considerably
smaller than the ambient magnetic noise in an urban environment, which is on the order of 108
fT or 10 T. The essential problem of biomagnetism is thus the weakness of the signal relative
to the sensitivity of the detectors, and to the competing environmental noise.
Figure 8: Magnetic eld of Brain
The MEG (and EEG) signals derive from the net e ect of ionic currents
owing in the
dendrites of neurons during synaptic transmission. In accordance with Maxwell's equations,
any electrical current will produce an orthogonally oriented magnetic eld(see gure 8). It
is this eld which is measured. The net currents can be thought of as electric dipoles, i.e.
currents with a position, orientation, and magnitude, but no spatial extent. According to the
right-hand rule, a current dipole gives rise to a magnetic eld that
ows around the axis of its
vector component.
13
5 f-NIRS
Functional near-infrared spectroscopy (fNIRs) has been introduced as a new neuroimaging
modality with which to conduct functional brain-imaging studies. fNIRs technology uses speci c
wavelengths of light, introduced at the scalp, to enable the noninvasive measurement of changes
in the relative ratios of deoxygenated hemoglobin (deoxy-Hb) and oxygenated hemoglobin (oxy-
Hb) during brain activity( see gure9). Wireless fNIRs system consists of personal digital
assistant (PDA) software controlling the sensor circuitry, reading, saving, and sending the
data via a wireless network. This technology allows the design of portable, safe, a ordable,
noninvasive, and minimally intrusive monitoring systems.
Figure 9: fNIR Device Working
The qualities of fNIRs make it an ideal candidate for monitoring cortical function in the
brain while subjects are engaged in various real life or experimental tasks. However, the noise
including in fNIRs is an important limitation on the use of optical data in these applications.
Motion artifact caused by moving of the head. Head movement can cause the NIR detectors to
shift and lose contact with the skin, exposing them to either ambient light or to light emitted
directly from the NIR sources or re
ected from the skin, rather than being re
ected from tissue
in regions of interest. These e ects cause sudden increases in the NIR data. Another noise can
cause the blood to move toward (or away from) the area that is being monitored, increasing
(or decreasing) the amount of oxygen, hence result in an increase (or decrease) in the measured
data. Hence, canceling noise from fNIRs signals is an important and necessary task in order to
deploy fNIRs as a brain monitoring technology in its full potential to many real life application
areas.
Adaptive ltering is one approach to dealing with noise signals. Adaptive ltering has been
widely used for noise reduction in other biomedical applications involving electrocardiogram
(ECG), EEG , and fNIRs.
5.1 Structure of fNIRs signals classi cation
Neural networks are very powerful tools for pattern recognition . The most well-known ad-
vantage is that after training them, neural networks can be readily used for process parameter
(or state) assessment without requiring any knowledge of the underlying system. In general, it
is necessary to preprocess their input information to eliminate irrelevant information from the
inputs and extract features of signals.
14
Figure 10: Structure of fNIRs signals classi cation
Here describe signal analysis to lter noises, feature extractions by wavelets techniques and
oine classi cation of the NIRS signal using Neural Networks. The structure of entire signals
processing is shown in Fig 10.
15
6 Conclusion
Measuring brain signals related to interfaces can lead to applications such as interface eval-
uation and adapta-tion. My thesis explores brain signals measured with fNIRS, use them to
adapt the interface and close the loop by connecting brain signals to the adaptable inter-face.
I am really enthusiastic about the potential for fNIRS and similar techniques to greatly en-
hance how people interact with computers. The creation of a brain-computer interface will
open opportunities for adapta-tion on di erent brain signals, with a device that is portable,
non-invasive and safe.
References
[1] \The future of brain-controlled devices". ACM, April 2010.
[2] \Audrey Girouard Computer Science Department Tufts University 161 College Ave Medford
USA audrey.girouard[at]tufts.edu". IBM J. Res. Dev.
[3] Pearlmutter B.A. Ward T.E. So-raghan C. Matthews, F. and Markham. \Hemodynamics
for Brain-Computer Interfaces". Signal Processing Magazine,IEEE, 2008.
[4] Leuthold AC Lewis SM Lynch JK Alonso AA Aslam Z Carpenter AF Georgopoulos A
Hemmy LS Koutlas IG Langheim FJ McCarten JR McPherson SE Pardo JV Pardo PJ
Parry GJ Rottunda SJ Segal BM Sponheim SR Stanwyck JJ Stephane M Westermeyer JJ
Georgopoulos AP, Karageorgiou E. \Synchronous neural interactions assessed by magne-
toencephalography: a functional biomarker for brain disorders". Signal Processing Maga-
zine,IEEE, Dec 2007.
