A
brain-computer interface (BCI) is a communication interface
between a human and a computer. BCIs provide a direct,
alternative means for humans to interact with machines
and vice versa. The communication medium allows one to interact
with the machine cognitively instead of through other means such
as tactile or verbal input. BCIs would therefore greatly benefit those
with severe disabilities such as locked-in syndrome or quadriplegia.
Locked-in syndrome is a condition in which a patient is aware and
awake but cannot move or communicate verbally due to complete
paralysis of nearly all voluntary muscles in the body except for the
eyes. BCIs would provide such individuals with a greater degree of
independence by enabling them to interact with their environment.
Spurred on by the possibility of assisting disabled people in this
way, students GrahamPeyton and Rudolf Hoehler from the University
of the Witwatersrand, Johannesburg, designed a brain-computer
interface that translates cognitive commands and facial gestures
into movements of a robotic arm. The students designed the system
as part of their fourth year Electrical Engineering project at Wits,
within the Biomedical Engineering Research Group in the School of
Electrical and Information Engineering, and under the supervision
of Adam Pantanowitz. Professor David Rubin is the leader of the
Biomedical Engineering Research Group, where researchers such
as Graham Peyton, Steven Dinger and Adam Pantanowitz have a
great interest in BCIs.
The system is a non-invasive BCI that makes use of electroen-
cephalography (EEG) and electromyography (EMG) to measure the
electrical activity produced by the brain on the surface of the scalp,
and the electrical activity of facial muscles responsible for facial
gestures. It is designed to enable the user to operate a robotic arm
in one of two ways.
Motor imagery
The first is to use ‘motor imagery’ – imagining or visualising a move-
ment of the robotic arm. In this case, EEG data are recorded from
the brain, and algorithms are used to detect and classify what action
that the user is visualising. The corresponding action is then sent to
the robotic arm.
The use of motor imagery, however, is less reliable than simply
mapping facial gestures to specific movements of the robotic arm.
For example, one may frown to move the robotic arm downwards.
Clenching one’s jaws, on the other hand, will close the pincers of the
robotic arm. It is possible to map one of the many facial expressions
or motor imagery to activities of the robotic arm – or by extension
of the principles – another physical device, communication device
or computer.
When the user performs a facial gesture, EEG and EMG data are
recorded simultaneously. Machine learning algorithms and signal
processing techniques are used to analyse and classify the electrical
signals that are produced when one carries out the facial expression.
This classification process enables a computer to determine the facial
expression of the person wearing the device from the measurements
received.
The system makes use of machine learning algorithms which
must be trained and optimised before it can be used. All the algorithms
were written in software using C++. Once an appropriate action has
been classified using the algorithms, the corresponding commands
are sent to the robotic arm. Training takes approximately 10 minutes
and is required for the system to respond to a specific individual’s
unique facial expressions. Training is done by firstly recording EEG
data under ‘neutral’ and ‘active’ conditions (such as frowning, clench-
ing, etc). The data is used to train the algorithm to classify facial
gestures based upon the characteristics of the EEG signals associated
with the corresponding gestures.
Brain switch
The system is also designed to address the problem of how severely
disabled people can turn the robotic arm on or off without the use of
their hands. The solution that Peyton and Hoehler came up with was
to implement an attention-based ‘brain switch’.
The switch exploits a physiological phenomenon called a Steady-
State Visually Evoked Potential (SSVEP). An SSVEP is a resonance
response produced in the brain when an individual gazes intently
at a light source flashing at a particular frequency. SSVEPs can be
detected in the occipital cortex of the brain using EEG. Algorithms
such as Power Spectrum Density Analysis (PSDA) and Canonical
Cross-Correlation Analysis (CCA) were used to detect the presence
of an SSVEP in the brain so as to identify whether the individual is
gazing at the light source or not. As the subject gazes at the light
source, an SSVEP is detected, the SSVEP ‘brain switch’ is turned on,
and the robotic armmay be operated. To turn the system off, the user
must then gaze at the light source again. Overall, this allows one to
operate the system with no tactile input.
I think - therefore I ‘Arm’
By G Peyton, Biomedical Engineering Research Group at the University of the Witwatersrand
The ability of computers to enhance and augment both mental and physical abilities is no longer the exclusive realm of science fiction. It is fast
becoming a reality. Brain-computer interface research is rapidly blurring the lines between man and machine.
Sensors, switches and transducers
Technology of the F tu e
Electricity+Control
December ‘13
36
