A LOW COST ERGONOMIC EEG SENSOR
FOR PREDICTING MENTAL ILLNESS
Dennis Majoe
1
, Jurg Gutknecht
1
and Hong Peng
2
1
Native Systems Group, ETH Zurich, Clausiusstrasse, Zurich, Switzerland
2
School of Information Science & Engineering, Lan Zhou University, Lan Zhou, China
Keywords: Mental Health, Depression, EEG Sensing, Wearable Smart Sensor.
Abstract: The EEG recording of a person has been considered as one potential component within an overall wearable
sensor system that predicts the onset of mental health problems. Such a smart EEG sensor should provide
detailed sensory information, be easy to use, and to put on and take off and whilst being very ergonomic the
design should aim at a very low final end user cost to ensure the widest possible take up by the e-Health
community. The work reported here describes the design of such a sensor, the performance and its use
during extensive clinical trials aimed to establish the rules that link physiology sensing to mental health
prediction.
1 INTRODUCTION
This work relates to a 5 electrode Electro
Encephalogram (EEG) sensor developed within the
EU research project OPTIMI. The project’s aim is to
provide on-line predictive tools for the early
identification and intervention during the onset of a
mental illness, in particular depression, following
the inadequate coping with day to day stress.
The EEG sensor is part of a set of wearable
sensors to be worn by a large number of volunteers
(130) during two phases of trials. The EEG sensor
will be used to measure Alpha band brain activity
and EEG signal C0 complexity. These
measurements will be calibrated and used in a data
fusion process to predict the onset of depression.
During the trials the EEG sensors must be used
every day for a period of around five minutes. In
order to ensure that volunteers comply with the daily
testing procedures, that will take place every day for
4 weeks, this daily activity must be as quick and
simple to perform as possible. At the same time high
EEG recording accuracy is necessary. If there are
difficulties or discomfort when using the sensor,
volunteers may drop out of the trials and the longer
term viability of the commercialisation of the sensor
will be left in doubt. If the sensor is not placed well
on the head the quality of the data may be
compromised. Therefore ergonomic design must be
considered at each stage of the development process.
Since a very large number of EEG sensors, as
high as 55 units, are required to meet the needs of
the trials, a low cost sensor is needed to ensure the
project keeps within its limited budget. Likewise, in
order to allow such technology to become
mainstream and used by large numbers of citizens,
low unit cost must be a central design criteria.
The sensor must provide equal data quality and
resilience to noise as the commercial off the shelf
EEG sensors that the OPTIMI psychologists currently
use in the laboratory and medical clinics and the
sensor data should be easily integrated into the
OPTIMI data processing systems that include
wearable Electrocardiograph and Actigraphy sensors.
This paper describes the need, usage, design,
development and testing of the EEG sensor covering
the electronics, software, ergonomics, economics and
accuracy.
Section 2 describes the OPTIMI architecture and
the EEG usage. Section 3 describes the development
of the sensor hardware and section 4 discusses the
results obtained during laboratory and the first phase
of OPTIMI trials.
2 SENSOR ARCHITECTURE
The OPTIMI project incorporates a number of smart
wearable sensors. The following summarizes the
sensors and their target function:
164
Majoe D., Gutknecht J. and Peng H..
A LOW COST ERGONOMIC EEG SENSOR FOR PREDICTING MENTAL ILLNESS.
DOI: 10.5220/0003732901640173
In Proceedings of the International Conference on Health Informatics (HEALTHINF-2012), pages 164-173
ISBN: 978-989-8425-88-1
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
EEG for stress and depression
Activity Sensor for ambulatory activity
ECG for Heart Rate and HR Variability
Sub Dermal Cortisol Level Sampler
Speech Analysis for depression score
The ECG and Activity sensors are worn 24/7
while the EEG sensor is to be worn once a day in the
evening for a few minutes in order to perform an
EEG recording and then removed. Data from all the
sensors is received wirelessly from each sensor by a
netbook computer called the HomePC. The HomePC
encrypts the data and transmits it to a central server
which processes the results.
Trials are to be conducted in Switzerland, Spain
and China over 4 weeks in order to collect
physiological data, volunteers’ self-reported data
and therapist interview data. By analysing the data
the plan is to derive heuristic rules linking
physiology to mental health prediction. When the
clinicians are satisfied with the rules the sensors will
be used as part of a Cognitive Behavioural Therapy
based on-line system. This system will be tested as
part of resilience treatment trials to be conducted in
the UK and Spain using the same sensors.
2.1 The Role of the EEG Sensor
The EEG is a record of the electrical potential
gradient oscillating around the brain and recorded
from electrodes on the human scalp and is often
labeled according to apparent frequency ranges
detected in the EEG signal power spectrum: delta (1-
4Hz), theta (4-8Hz), alpha (8-13Hz), beta (13-
20Hz), and gamma (roughly >20Hz). The scalp
electric potential amplitude is typically 20 to 100 μV
with a specific shape depending on the subject’s
state of relaxation (Srinivasan, 2006). Recent studies
have suggested that alpha rhythm is an oscillatory
component of the human EEG and applied widely in
many application fields, such as mental illness,
biometric identification, E-learning, etc. Both wake
and sleep EEG can provide biomarkers of depression
and anti-depressive therapy, respectively (Tang,
2009). Within OPTIMI the resting EEG is recorded
at three points on the forehead (FP1, FP2 and FPZ)
referenced to the two ear lobes (A1 and A2).
The majority of studies targeting depression
report characteristic differences in EEG asymmetry,
especially in the alpha band, recorded from
electrodes placed on frontal locations. In particular
the study performed by Miguel A. Diego supports
previous findings that indicate that greater relative
left frontal EEG alpha activity is evident in people
with depressive symptomology (Diego et al., 2002).
In addition the findings of Vuga et al support the
view that resting frontal EEG asymmetry reflects a
moderately stable individual difference in adults,
irrespective of sex and history of depression. [4]
Another good candidate for predicting
depression is the measurement of EEG signal
complexity. J L Nandrino et al. found that the EEG
dynamics of major depressive subjects is more
predictable, that is less complex, than that of control
subjects(Nandrino et al., 1994). Tang et al. showed
that alpha rhythm entropy in depressives is increased
during resting compared to when perfroming mental
arithmetic and when compared to healthy control
group (Tang, 2009).
Within OPTIMI the two above measures, Alpha
asymmetry and C0 Complexity, will be of particular
interest as well as Beta/Alpha and Alpha/Theta power
band ratios which can reflect a person’s mood.
The EEG sensor must reliably record the user’s
EEG signals over periods of approximately 90
seconds in order to obtain sufficient data for the
above algorithms. To achieve this the sensor must
meet certain electrical specifications which will be
described later. Additionally the sensor should
provide a means to grade the quality of contact
between the electrodes and the skin on the forehead
since this contact quality must be sufficient if the
raw data is to be useful.
Artefacts due to eye movements and forehead
muscle movements are often detected in EEG raw
data. Both 50Hz noise from electrical power lines
and artefacts must be removed by a denoising
algorithm to provide pure data before the algorithms
are applied.
At first the algorithms and de-noising will be
computed on the HomePC with the sensor purely
acting as a wireless EEG recorder controlled by the
HomePC. However once the results of the
calibration trials are fully interpreted the aim is to
move many of these data processing tasks to the
sensor, including a neuro feedback mode, making it
less platform dependent and highly portable.
3 SENSOR HARDWARE
The sensor electronics is depicted in Figure 1. The 5
electrodes on the left are connected first to a signal
conditioning circuit to clamp any voltage spikes that
may arise due to electrostatic discharge. In addition
passive filtering is performed to remove 100Hz or
higher input noise.
The electrodes connect to A1, the left ear lobe,
FP1, the left hand side of the mid forehead, FP Zero,
A LOW COST ERGONOMIC EEG SENSOR FOR PREDICTING MENTAL ILLNESS
165
Figure 1: EEG Sensor, primary hardware components.
the middle of the forehead, FP2 the right hand side
of the middle forehead and A2 the right earlobe.
These signals are used to drive the 8 channels of
the ADS 1298 instrumentation amplifier ADC from
Texas Instruments (Texas Instruments). This device
provides 8 high impedance separate inputs for multi-
channel amplification and multiplexing and digitizing
any selected channel to a resolution of 24 bits.
The 5 input leads are inter-connected in a manner
so as to obtain the following channel combinations:-
FP1 relative to A1
FP0 relative to A1
FP2 relative to A1
FP1 relative to A2
FP2 relative to A2
FP1 relative to FP2
At the centre of the sensor is a STM32 F101CB,
(from ST Microelectronics) (ST Microelectronics )
which is an ARM 32-bit Cortex™-M3 CPU with a
maximum 36 MHz clock, 1.25 DMIPS/MHz
(Dhrystone 2.1) performance. It includes 128 Kbytes
of Flash memory, 16 Kbytes of SRAM, two SPI data
buses and a generous number of general purpose I/O
lines.
The CPU communicates with the ADS 1298 via
a dedicated SPI bus configured to operate at the
highest speed available. The second SPI bus is
connected to a block of flash memory and an RF
communications front end.
The 32MB of flash memory provides the main
storage for the sensor. The amount of memory is
only restricted by PCB real estate and memory cost.
The RF frontend allows the CPU to accept
commands from the HomePC as well as exchange
EEG raw data or processed results. In order to
maintain compatibility with other sensors in the
OPTIMI project, the RF front end is based on the
nRF24L01+ low power 2.4GHz ISM (Industrial,
Scientific and Medical) band RF Transceiver from
Nordic Semiconductor (Nordic Semiconductor WEB
Site). This method of communications was preferred
over traditional methods such as Bluetooth because
the hardware platform is extremely accessible and
allowed one to develop energy efficient
communications protocols.
The standard EEG channel specifications
required by such typical applications are mentioned
in the review of Tan, Ibrahim and Moghavvemi (Tan
et al., 2007) in which they suggest:-
Input impedance: 47Mohms
Bandwidth: 1 to 30Hz.
CMRR: 80dB to 90dB
Gain: 100,000
Power: 3mW per channel
Commercial products such as the Nexus-4 (from
Mindmedia) (Mind Media, WEB Site) and the
ENOBIO (from Starlabs) (Starlabs WEB Site) quote
similar or better figures.
The ADS1298 is derived from a family of
multichannel, simultaneous sampling, 24-bit, delta-
sigma analogue-to-digital converters with built-in
programmable gain amplifiers, internal reference,
and an on-board oscillator. This component has an
extremely low input bias current of 200pA (typical)
and an input-referred noise of 4μVPP (typical). The
CMRR is about -115dB, 500Megohm input
impedance, effective gain of 1,000,000 and the
power is very low in the order of 0.75mW/Channel.
The data rate is from 250SPS to 32kSPS. The chip is
9mm x 9mm and requires few external components.
Therefore a very compact design may be achieved.
HEALTHINF 2012 - International Conference on Health Informatics
166
In this sensor, the gain is set to 6, the reference
voltage at 2.4V and the sampling rate is 260Hz
which can also be easily set to other rates if
necessary.
The sensor incorporates a 470mAh Lithium
Polymer single cell battery. This allows the sensor to
be used continuously for approximately 6 hours.
Since the trials are planned to include twenty four
sessions each lasting 5 minutes a single full charge
should ensure the sensor never needs to be recharged
during the whole trial period.
3.1 Firmware Functionality
The sensor software currently comprises a number
of critical firmware functions. These include:-
Accepting commands from the HomePC over
the wireless communications link.
Performing a signal quality check and relaying
this to the HomePC
Recording a specified length of EEG data into a
simple multiple file structure
Streaming raw EEG data for test purposes
Downloading data files to the HomePC over the
RF link
Erasing data files
An important feature of the sensor is the ability
to perform a signal quality check. This feature is
necessary to provide the HomePC application a
means of verifying that the user has put the sensor
onto their head, in a manner that ensures the data
recorded is true EEG data with little noise.
When any EEG sensor is used to record very low
voltage brain waves activity, care must be taken to
ensure the electrodes are in good contact with the
skin such that the electrode to skin impedance is as
low as possible. This ensures an optimal detection of
the brain potentials. In addition it ensures that any
ambient electrical potential, such as generated by
nearby electrical lamps and office machines is less
likely to interfere due to the fact the electrodes
measure the potential originating in the human body
rather than the potential gradient in air space.
The signal quality check is performed in a fully
standalone manner by the sensor following the
reception of a GetSignalQuality command by the
HomePC. It first records 4 seconds of EEG data and
then carries out a detailed analysis in which the
sensor determines the Variance in the time series
data as well a power spectral analysis.
A value for signal quality is calculated based
upon:
the variance in the signal which should be
between a set of typical limits
the spectral power at 50/60 Hz
the brain activity in the Alpha band
the brain activity in the Delta, Alpha and
Beta bands as compared to the Gamma and
higher frequencies to 60Hz.
By a combination of ratios and limits a signal
quality value is calculated between 0 and 100 and
this value is sent to the HomePC which instructs the
user to check or reposition the sensor if the quality
falls below 70.
3.2 Choice of Electrodes
In recent years a great deal of work has been applied
to developing active dry EEG electrode technologies
both in the research and commercial domains [12.
The aim has been to achieve a number of advantages
over the traditional approach. In the traditional
approach a conductor such as a stainless steel or gold
plated electrode is brought near the surface of the
skin. Then a wet gel such as NUPREP™ EEG skin
prepping gel is used to lower the skin conductance.
The gel is rather messy and is particularly annoying if
it gets into ones hair. Normally the electrodes are
connected to relatively long and heavily shielded
expensive cables which go off towards the amplifier
device next to a patient’s bed.
As a result research teams have aimed to place
active electronics into the electrodes so that they
could pre-amplify and match impedance close to the
skin and remove the need for a gel as well as avoid
the need for active shielded cables.
In the commercial solutions now available and
known to the authors, it would appear that active
electrodes result in a more expensive design due to
the higher complexity, additional amplifiers and
power. The electrodes are also larger than the
smallest passive electrodes.
For economic and ergonomic reasons the sensor
developed here did not incorporate active electrodes
and instead makes use of very low cost disposable
skin friendly solid gel pads. Solid gel pads provide
all the electrical benefits of wet gel and do not leave
any gel or fluid on the skin which is often the
complaint during EEG usage. When the recording is
completed the pads are simply removed and
discarded, providing a higher level of hygiene.
3.3 Ergonomics and Economics
From the very start of the OPTIMI sensor
development, two things were made very clear to the
designers. Firstly, the EEG sensor’s final assembled
A LOW COST ERGONOMIC EEG SENSOR FOR PREDICTING MENTAL ILLNESS
167
and tested unit cost would have to be below 90
Euros. Secondly the design would have to pass a
usability acceptance rate of 39 out of 40 users
invited to use it as deployed in the trials. That meant
that only 1 in 40 subjects should be likely to refuse
to use the sensor on ergonomic grounds. This set
extremely difficult economic and ergonomic
constraints in addition to the normal functional
design constraints of the sensor.
Economics
In order to achieve low cost the designers reviewed
the state of the art and determined different cost
scenarios for alternative designs, manufacturing
methods, battery charging methods, user interface
and packaging. A core base functionality was
necessary and was achieved through the value line
STM32 F101 device, a limited 32MB of serial flash
memory and the compatible nRF24L01+ RF
transceiver.
After studying different options it became clear
that the single ADS 1298 device, encapsulating all
the functionality of the EEG front end and ADC,
would save money over the use of discrete
components and the associated assembly cost.
Active dry electrode sensors were considered in
the design. However since it became clear that the
sensor would be worn on the forehead, with very
short electrode cables there was no technical reason
to implement this higher cost solution. In addition it
was decided that disposable solid gel pads, which
cost a few cents each, were to be used to optimise
skin conductance while maximising hygiene.
Battery charging is performed using a non-
contact inductive method. This is done to maintain
compatibility with all OPTIMI sensors which must
be hermetically sealed, as they are worn for long
times near the body.
The final packaging of the device is kept very
simple, with the overall PCB being coated in an
encapsulating two part epoxy resin (ALH Systems
Ltd., U.K.) that has been chosen to provide
maximum water resistance and maintain hardness to
over 80 degrees C. The encapsulated board is then
primed and coated with a thin conductive coating
paint and then varnished. This conductive layer
helps to reduce the effects of ambient electrical
potentials being detected. In fact due to the small
size of the final sensor and direct coupling of the
ADS 1298 to the shielded electrode cables, the
effects of ambient noise are surprisingly small.
The sensor is worn on a low cost head band
which was chosen for economic but also strong
ergonomic reasons.
Ergonomics
To assist in the design of an ergonomic sensor a
usability study was commissioned to run in parallel
with the R&D work. This study consisted of a
developers’ session, two focus group sessions and a
detailed survey conducted in each of the trial sites.
To start with the sensor design and developers
group were invited to highlight concerns and
constraints of developing these sensors. Then in the
first focus group session the OPTIMI philosophy
and mock up sensors were presented to 14
volunteers 7 male, 7 female with age ranges from 18
to 64.
Figure 2: Early mock up of the EEG sensor.
By studying the response to the mock ups and
scenarios of use, the concerns of the developers were
matched against the concerns of the focus group to
highlight important design goals and constraints.
Following this, design specifications were
created and prototype work commenced. During the
second focus group, working prototypes were
presented to 18 volunteers who were asked to wear
and use the sensors. As a result further feedback was
obtained, prioritised and several design changes
resulted.
Finally questionnaires were sent out to 50 people
representing typical volunteers in each trial site.
These questionnaires further clarified if there were
significant cultural adjustments that needed to be
made given that the focus groups were held only in
Switzerland.
As a result of this usability study a large number
of design modifications were made to the developers
original design ideas. In the end it was felt the
design satisfied the specific needs of the people in
the focus groups as well as the points derived from
the regional surveys. The following describes a few
of the important conclusions coming from this
usability study:
HEALTHINF 2012 - International Conference on Health Informatics
168
Figure 3: EEG sensor, from front, from behind. Note the electrodes are carried on the black elastic headband. The weight of
the electronics is supported on a solid band worn as eyeglasses.
Figure 4: Sensor worn by a volunteer.
10 seconds to put on or remove
Quick Recharge every 1 to 2 months
All cables must be short and ultra-flexible
Very light pressure, light weight ear clips
Light weight headband and small electrodes
Electronics separated from the band
High preference for disposable solid gel
pads
No skin allergy reaction
Use soft feel hook and loop straps
Lady’s long hair must never get caught
LEDs to tell what the sensor is doing
Simplest HomePC application
4 PERFORMANCE RESULTS
The EEG sensors developed and finally
manufactured have recently begun deployment in
large numbers (45 currently) in the calibration trials
in Spain, Switzerland and China.
Before this the sensors were rigorously tested
and data from the sensors was used in laboratory
experiments to ensure the EEG data obtained was of
the highest quality for use in the trials.
For this paper the performance will be described
by first showing the time and frequency domain
results obtained under different conditions. Then a
comparison will be presented between the OPTIMI
EEG sensor and a well-known commercial product.
Finally some initial results obtained from the on-
going calibration trials will be used to indicate the
ergonomic performance and acceptance by users.
Time and Frequency Domain
In order to compare the sensor under different noise
conditions recordings were made of volunteers in the
A LOW COST ERGONOMIC EEG SENSOR FOR PREDICTING MENTAL ILLNESS
169
resting, eyes closed position.
Two environments were tested. First the
recordings were taken in a public wilderness park
near Zurich where there are no visible sources of
electrical power usage. With several volunteers to
record, this was simpler, faster and less costly
exercise when compared to using a dedicated Faraday
cage system. Secondly a typical office environment
was arranged with several running desk top
computers, lamps, printers and copiers in the room.
Raw EEG data was gathered using the sensor
alongside a PC based test application software. The
raw data was then processed using MATLAB for
visualisation and for the time and frequency domain
analysis (TD and FD).
Analysis of Raw Data
Figures 5 to 12 relate to the recorded signal from a
single channel (FP1 relative to A1). Figure 5 shows
the TD signal with the sensor left to rest in free
space and not placed on the head of the volunteer.
Figure 5: Forest, Off head, Noise, Time Domain.
The peak amplitude is 9uV with an RMS value
of 4uV. This noise includes white noise in the
environment as well as sensor produced noise.
Figure 6: Forest, Off head, Noise, Frequency Domain.
In Figure 6, the FD power spectrum for the
signal is shown. In general the noise power is
distributed over all frequencies, with some
amplification at around 8Hz with no frequency
exceeding 0.6uV.
In Figure 7, the sensor is worn on the head, in the
forest environment. The volunteer keeps their eyes
closed and rests in a calm state of mind and body. The
primary aim is to verify the typical EEG pattern in a
noise free environment. The signal peaks at 60uV
with a typical RMS value of 25uV. We assume 4uV
RMS of noise is embedded in this signal.
Figure 7: Forest, On head, Resting Eyes Closed, Time
Domain.
Figure 8 shows an excellent and typical noise
free EEG resting eyes closed FD spectral plot. With
the eyes closed the FD plot clearly shows the Alpha
Band peaking around 10Hz, with additional Theta
band activity and low Beta band activity.
Figure 8: Forest, On head, Resting Eyes Closed,
Frequency Domain.
In the office with the sensor left to rest in free
space, the sensor records airborne electrical potential
gradients from the office machines. Figure 9 shows
the TD signal, peaking at 200uV with a typical RMS
of 170uV. Figure 10 shows the FD plot clearly
indicating that the major energy is detected at the
50Hz band.
In the office with the sensor left to rest in free
space, the sensor records airborne electrical potential
gradients from the office machines. Figure 11 shows
the TD signal, peaking at 200uV with a typical RMS
HEALTHINF 2012 - International Conference on Health Informatics
170
of 170uV. Figure 7 shows the FD plot clearly
indicating that the major energy is detected at the
50Hz band.
Figure 9: Office, Off head, Noise, Time Domain.
Figure 10: Office, Off head, Noise, Frequency Domain.
Figure 11: Office, On head, Resting Eyes Closed, Time
Domain.
Figure 11 and 12 show the results of recording in
the office with the sensor worn correctly in the eyes
closed resting position. What is now clear is that the
office noise at around 50Hz is very apparent within
the EEG FD plot. However the noise in the office
has not in any way compromised the response at the
frequencies of interest namely the Alpha and Theta
bands. The 50 Hz noise measured here is largely due
to the potential gradient set up in the human body
and measured between the two electrodes. These
gradients are due to tiny electrical currents arising
from the capacitively coupled voltages on equipment
in the nearby environment entering the body and
passing through it towards the earth or another
nearby equipment. Since these are truly human body
voltages there is no real method to supress them
other than to add a 50Hz band stop filter on the
processing stage of the raw data.
Figure 12: Office, On head, Resting Eyes Closed,
Frequency Domain.
Comparison with Commercial Eeg
In order to determine if the sensor was performing as
well as an off the shelf commercial sensor, a test was
carried out in which a volunteer wore both the OPTIMI
sensor and a NEXUS-4 EEG sensor from Mindmedia
(Mind Media, WEB Site). The NEXUS-4 device is a
very popular wireless Bluetooth based device used in
many research groups and clinical settings and has an
impressive specification. It is rather larger than the
OPTIMI sensor and cannot be worn as ergonomically.
It is limited to 4 electrodes and would be harder to
integrate seamlessly into the OPTIMI wireless system.
It is priced at around 3000 Euros.
Figure 13 shows the volunteer wearing the
OPTIMI EEG headband on her forehead. Above the
band can be seen the single electrode from the
NEXUS-4 EEG sensor stuck near the OPTIMIT
EEG sensor electrodes. Lower down one can see the
NEXUS-4 processing unit being held in her hand.
Figure 14 to 16 show the time domain and
frequency domain results comparing the signals
recorded from the two sensors. In Figure 14 and 15
is shown the time domain raw data plot. In the first
figure one can see that the general amplitude, DC
drift and general shape of the two waveforms are
more or less identical. On close inspection one can
see that the phase of the two signals is very
synchronised while the amplitude in some cases
varies significantly. Since the two sensors do not
share the exact same physical location and
A LOW COST ERGONOMIC EEG SENSOR FOR PREDICTING MENTAL ILLNESS
171
impedance contact with the skin, some differences
are likely to occur.
Figure 13: Comparison of NEXUS-4 and OPTIMI EEG
Sensor.
Figure 14: Drift comparison with NEXUS-4.
Figure 15: Phase Amplitude Comparison with NEXUS-4.
From the frequency response shown in Figure 16,
one can see that both sensors manage to provide
similar
FD plots, especially highlighting the peaks in
Alpha, Theta and Beta bands. The amplitude across the
spectrum is very close suggesting very similar input
response functions.
Figure 16: Frequency Response Comparison with
NEXUS-4.
Ergonomics and User Acceptance
Approximately 45 EEG sensors are now deployed in
three trial sites around the world. In certain cases the
sensor is shared between two or more volunteers in
dormitory scenarios. Over the past 5 weeks 52
volunteers have now used the sensor as required. We
can report that to date not a single user has found
any difficulty in putting on the sensor and
performing a recording. Some difficulties have been
experienced in operating the HomePC application to
start and end the recording. In a specific case of our
experiments, in which audio clips were played to
volunteers during the recording period, the
volunteers were required to wear headphones at the
same time. This lead to the user dislodging the
reference ear clip from the ear lobe and resulting
loss of data. As a result adhesive dry gel pads are
used behind the ear at M1 and M2. In some cases the
ear clip itself was broken as users pulled on the
cables to remove the electrode. The ear clip has
therefore been redesigned.
5 FURTHER WORK
The current version of EEG sensor will be modified
to include the noise and eye movement filtering that
is established post calibration trials.
In addition it is planned to add a neurological
feedback functionality in which the sensor will
identify the current level of mental stress of the user.
This level will be presented on a simple level
HEALTHINF 2012 - International Conference on Health Informatics
172
indicator, such as an audio signal. The sensor may
then be used as a simple assistive tool to practice
Cognitive Behavioural Therapy based biofeedback
aimed at relaxation before going to sleep.
The sensor will also be used as a low cost
diagnostic tool by project partners to evaluate other
forms of assistive diagnosis of stress and anxiety.
6 CONCLUSIONS
During the first year of the OPTIMI project a low
cost ergonomic sensor has been developed and
manufactured, from the bottom up, to meet the needs
of a large e-health trial.
The need to ensure volunteer compliance and
tight economic controls has been met to a high
degree in the work reported. This has been achieved
by designing the sensor in parallel with the guidance
of a user needs analysis that includes the views of
developers, the views of user represented by focus
groups and the views of typical end users in the
country of deployment.
Economic targets have been met by using state of
the art solutions combined with sophisticated
embedded software. The sensor is now in use in high
numbers and will soon be adapted to meet the needs
of customised usage scenarios by firmware upgrade.
ACKNOWLEDGEMENTS
We acknowledge the co-funding received under the
EU project OPTIMI www.optimiproject.eu .
REFERENCES
Paul L. Nunez Ramesh Srinivasan, Electric fields of the
brain. The Neurophysics of EEG, OUP US (2006).
Tang Y., Li Y, Tong S., Li Y, Zhu Y .Entropy analysis of
the EEG alpha activity in depression patients. Sheng
Wu Yi Xue Gong Cheng Xue Za Zhi,26(2009),739-742.
Miguel A. Diego B. A., Tiffany Field Ph.D, Maria
Hernandez-Reif Ph.D., " CES-D depression scores are
correlated with frontal EEG alpha asymmetry.",
Depression and Anxiety.Volume 13, Issue 1(2001),32–
37., Physicians Postgraduate Press, January 2002,
pages 6-31.
Vuga, M., Fox, N., A., Cohn, J. F., George, C., J.,
Levenstein, R., M.,Kovacs, M. Long-term stability of
frontal electroencephalographic asymmetry in adults
with a history of depression and controls,
International Journal of Psychophysiology
59(2006),107–115.
Nandrino JL, Pezard L, Martinerie J, el Massioui F,
Renault B, Jouvent R, Allilaire JF, Widlöcher D.
Decrease of complexity in EEG as a symptom of
depression. Neuroreport. Jan 12;5(4)(1994),528-30.
Texas Instruments WEB Site Analogue Front End
http://www.ti.com/ww/en/analog/ads1298/.
ST Microelectronics WEB Site Microcontrollers
http://www.st.com/internet/mcu/class/1734.jsp
Nordic Semiconductor WEB Site RF 2.4GHz
http://www.nordicsemi.com/eng/Products/2.4GHz-
RF/nRF24L01P
Tan, Y., Ibrahim, F., Moghavvemi, M., “Two electrodes
low voltage input EEG amplifier for brain-computer-
interface”, International Conference on Intelligent and
Advanced Systems, Kuala Lumpur, 2007.
Mind Media, WEB Site, Nexus-4 http://www.mindmedia.
nl /english/nexus4.php.
Starlabs WEB Site, http://starlab.es/products/enobio
Fonseca, C., Cunha, J., Martins, R., Ferreira, V. M., de Sa,
J., Barbosa, M., da Silva, A. M., “A Novel Dry Active
Electrode for EEG Recording”, IEEE Transactions on
Biomedical Engineering, Jan. 2007, Volume: 54
Issue:1 162 – 165.
A LOW COST ERGONOMIC EEG SENSOR FOR PREDICTING MENTAL ILLNESS
173