Motor Systems
NeuroReport 10, 1767±1771 (1999)
THE feasibility of using the EEG signal to operate a
hand grasp neuroprosthesis was investigated. Two able-
bodied subjects and one neuroprosthesis user were
trained to control the amplitude of the beta rhythm
recorded over the frontal areas. After 6 months, all
subjects exhibited a high level of control, being able to
use this signal to move a cursor to targets on a
computer screen with a high (. 90%) accuracy rate.
Control over the EEG signal was unaffected by upper
extremity movement or electrical activation of the
muscles, indicating that this signal would be adequate
for neuroprosthetic use. To test this concept, the
neuroprosthesis user operated his system with the
cortical signal, and was able to effectively manipulate
several objects. NeuroReport 10:1767±1771 # 1999 Lip-
pincott Williams & Wilkins.
Key words: Command±control; Electroencephalogram;
Neuroprosthesis
Introduction
Functional neuromuscular stimulation (FNS) has
restored functional hand grasp in individuals who
have sustained a spinal cord injury at the ®fth or
sixth cervical level. The current hand grasp neuro-
prosthesis developed at Case Western Reserve Uni-
versity and the Cleveland VA Medical Center [1±5]
electrically excites the muscles of the forearm and
hand to provide the user with two possible grasp
patterns. The control over the degree of hand open-
ing and closing is provided by either an external [5]
or implanted [6] transducer mounted at the shoulder
or wrist. However, as the neuroprosthesis continues
to develop to allow its implementation in both arms
or implantation in persons with a higher cervical
level injury, the need for other methods of control-
ling the system becomes greater.
There are currently several options for the control
of the neuroprosthesis besides the use of shoulder or
wrist motion. These include muscle activation [1],
head movement [7], respiration control [8] and voice
control [9]. Another method which has been pro-
posed is the use of signals recorded through the use
of intracortical electrodes. Kennedy et al. [10,11]
have successfully developed and implanted an elec-
trode into the cortex of an individual with amyo-
trophic lateral sclerosis (ALS) to allow him to
operate a computer cursor. These signals might have
future applications in controlling electrically stimu-
lated muscles.
Other investigators [12±14] have also studied the
signals which can be acquired from the neurons of
0959-4965 # Lippincott Williams & Wilkins
NeuroReport
EEG-based control of a
hand grasp
neuroprosthesis
Richard T. Lauer,CA P. Hunter Peckham
and Kevin L. Kilgore
Case Western Reserve University, FES Center of
Excellence, Cleveland VA Medical Center,
Hamann 601, MetroHealth Medical Center, 2500
MetroHealth Drive, Cleveland, OH 44109, USA
CA
Corresponding Author
the motor cortex. They observed that a neuron in
the motor cortex will ®re in response to a given
direction or muscle action, and by recording from
these neurons, it is possible to reconstruct the
desired movement. One goal of this research, still in
the early fundamental research stage, would be to
use this complex signal to operate a neuroprosthesis.
The use of intracortical signals to operate the
neuroprosthesis is attractive since it would allow for
a more natural operation of the hand by restoring
the link between the brain and hand movements.
Another possible method of using the brain signals
to operate the neuroprosthesis is the EEG recorded
from the scalp using surface mounted electrodes.
This has several advantages over the use of the
intracortical recordings. First, it is non-invasive and
second, it would allow for a more rapid implementa-
tion into the current neuroprosthesis. Wolpaw et al.
[15,16] and Pfurtscheller et al. [17] have demon-
strated that subjects can be trained to voluntarily
control the amplitude of a speci®c frequency com-
ponent of the EEG. This signal has been used to
move a cursor on a computer screen, which is the
basis of an assistive communication device for those
persons with severe physical disabilities (i.e. ALS
and stroke).
Birbaumer et al. [18,19] have also investigated the
use of cortical signals recorded from surface
mounted electrodes. However, their studies differ
from those of Wolpaw and Pfurtscheller in that the
focus is on the self-regulation of slow cortical
potentials instead of a speci®c frequency component
of the EEG. Birbaumer and his group have success-
Vol 10 No 8 3 June 1999 1767
NeuroReport
fully developed a thought translation device (TDD)
which allows individuals in a locked-in state to
communicate by selecting letters with the slow
cortical potential to form words.
The objective of this study was to develop an
EEG-based controller for use with the neuroprosth-
esis based upon the system developed by Wolpaw.
To investigate the feasibility of this approach, three
studies were conducted: training of both able-bodied
subjects and neuroprosthesis users to control the
beta rhythm recorded from the frontal cortex;
evaluating interference with the EEG control signal
introduced by factors such as extremity movement
and artifacts of electrical stimulation; and developing
an interface between the EEG signal and the neuro-
prosthesis.
Materials and Methods
Subjects: Three subjects (two able-bodied and one
neuroprosthesis user) ranging in age from 27 to 40
participated in these studies. The neuroprosthesis
user was male, while in the able-bodied group, one
was male and one was female. All subjects were
asked to read and sign an informed consent form
which was approved by the Institutional Review
Board of MetroHealth Medical Center before parti-
cipation in the study.
Beta rhythm control: The instrumentation and pro-
tocols used for the training of the subjects to control
the amplitude of the beta rhythm were similar to
those used by Wolpaw et al. to control the mu
rhythm [12]. Only a preliminary description of the
methods and of the brain±computer interface (BCI)
are provided since these have been covered exten-
sively in previous publications [20,21].
The subject was seated in an upright position
facing a computer monitor upon which appeared
targets at the top and bottom of the screen and a
cursor which they could control. Sixty-four electro-
des mounted in a cloth cap (Electro-Cap Interna-
tional, Inc.) were placed on the scalp, the spacing
and location based on a modi®ed 10-20 system. The
output from each of the electrodes was referenced to
the right ear, and the ground was located behind the
ear on the mastoid process. Referential recordings
are used in this case to maximize the amplitude of
the EEG signals being recorded. The signal recorded
from each of the electrodes was ampli®ed (320 000)
and bandpass ®ltered (1±35 Hz). The output from
the ampli®er was sent to two computers. One
computer stored the raw EEG signal for later off-
line analysis. The second computer was responsible
for the conversion of the EEG signal into cursor
movement. From the 64 electrodes, a subset of 5 (or
1768 Vol 10 No 8 3 June 1999
R. T. Lauer, P. H. Peckham and K. L. Kilgore
10) was used to generate cursor movement. One (or
two) of the electrodes constituted the recording sites
used to drive cursor movement, while the remaining
four (or eight) electrodes were used to provide
spatial ®ltering. The recording sites used for each of
the subjects are presented in Table 1.
Subjects were trained to control the amplitude of
the beta rhythm. The beta is the 18±40 Hz compo-
nent of the EEG signal recorded primarily from the
frontal and somatomotor cortices. For the subjects
in this study, the beta band was restricted to be-
tween 25 and 28 Hz. Training for each subject
involved 1±3 sessions per week. Each session con-
sisted of eight runs which were 3 min in duration
with a 1 min break between runs. During the course
of a given run the subject was presented with 30±35
targets generated randomly, depending upon the
skill and speed of the subject. The initial sessions
were used to determine what mental states caused
movement of the cursor up and down. These usually
involved the subject focusing upon the cursor or
target to make the cursor move up, and relaxing to
make the cursor move down. As training progressed,
subjects were able to move the cursor by thinking of
the appropriate direction.
Neuroprosthetic issues: The second stage of the
study involved addressing issues relevant to using
the EEG to operate the neuroprosthesis: can the
subject control the EEG signal while generating
limited movements of the upper extremity; and will
the stimulation of the muscles provided by the
neuroprosthesis interfere with the operation of cur-
sor movement?
To address the ®rst question, each able-bodied
subject was asked to perform the following experi-
ment. The subject was seated in front of the
computer monitor, as before. However, on the table
were also placed a white line which corresponded
with the midline of the subject and a 0.5 kg weight.
The session began with the subject moving the
cursor to targets appearing on the computer screen.
However, the length of the run was decreased to
2 min and the targets appeared in a ®xed pattern.
After completing two runs, the subject was then
instructed to reach out for the weight and grasp it
with the right hand, move the weight across the line,
Table 1. Location of the recording sites for the frontal beta
rhythm. The designations are based on the modi®ed 10-20
system. The beta rhythm was recorded from two sites in subject
AB-1. In the other subjects, only one site was used
Subject Recording site
AB-1 F3 and F4
AB-2 FP1
NP-1 F3
EEG-based control of a hand grasp neuroprosthesis
release the weight and return their hand to their lap.
They were to continue to do this at a self pace that
was comfortable while moving the cursor to targets
on the screen. This continued for the entire 2 min
run. In the next run, the subject was instructed to
repeat the same movement with the left hand. The
series of non-movement, right side movement, and
left side movement was then repeated twice more,
for a total of nine runs. For the neuroprosthesis
user, this protocol was modi®ed slightly since he
was unable to grasp the weight and move it when
his system was off, but could move the proximal
joints (i.e. shoulder and elbow). Therefore, move-
ment was restricted to reaching out and touching
the weight. All other aspects of the study were the
same.
To address the second issue, the neuroprosthesis
user was asked to turn his system on and lock his
hand in an open or closed position, and then
proceeded with a normal training session. The
system was turned off between runs in order to
prevent fatigue of the muscles.
Neuroprosthetic control: Neuroprosthetic operation
with the EEG signal was implemented to demon-
strate the feasibility of this control method. The
BCI system was modi®ed to allow the signal which
is used to drive the cursor movement to be sent out
of the computer through the serial port into a
second computer. The second computer, running a
routine written using the LabVIEW programming
language, provided additional signal processing and
conversion of the EEG signal into a command for
the neuroprosthesis. The experimental setup for the
hybrid system in shown in Fig. 1.
The EEG signal was converted into a command
signal using the gated ramp method. This method
has been used previously [7] for the conversion of
the myoelectric signal into a command for the
PC EEG
signal amplifier
processing Electrode
matrix
PC
signal Display
Implanted
conversion stimulator
External
control
unit
FIG. 1. Schematic showing the components of the EEG-based control-
ler for the FNS hand grasp system. This system allowed for the
conversion of the raw EEG signal into a command-control signal for the
neuroprosthesis to provide the subject with hand opening and closing by
thinking about it.
NeuroReport
neuroprosthesis. When the signal was maintained
above a preset threshold, this generated the com-
mand to go from hand opening to hand closing. The
rate at which the hand closed was ®xed (,2±3 s to
full closure), and was only generated when the
amplitude of the input signal was maintained above
the threshold. When the signal fell below the thresh-
old, the hand stopped closing. To go from the hand
closed to the hand open position, the signal must be
below a preset low threshold, which generated the
command to go from hand closed to hand open at a
much faster rate than hand opening (,1 s to full
opening).
The use of the gated ramp only allowed for
dynamic hand operation. Therefore, the subject was
only asked to perform simple activities of daily
living with the system. These tasks included picking
up and moving a weight, grasping and releasing a
fork, and grasping and releasing a cup. The subject
was also asked to open and close his hand using the
EEG signal as quickly as possible, and at the verbal
command of the investigator. Finally, the subject
was also asked to open and close his hand in the
absence of visual feedback (i.e. not looking at the
hand to determine whether it was open or closed).
Results
Beta rhythm control: The ability of the subjects to
control the amplitude of the beta rhythm, as meas-
ured in accuracy rate, is shown in Fig. 2. The
subjects accuracy rate is plotted as a function of time
(training session). The neuroprosthesis user is given
the designation NP-1, while the two able bodied
subjects are given the designations AB-1 and AB-2.
During the 6-month period the subjects participated
in 10±20 training sessions. By the end of this time,
all three subjects were able to achieve excellent
control over the beta rhythm, achieving on a con-
sistent basis an accuracy rate . 90%. Subject NP-1
100
95
90
Accuracy (%)
85
80 AB-1
75 AB-2
70
NP-1
65
60
55
50
0 5 10 15 20
Session
FIG. 2. Plot of subject accuracy over time. Accuracy rate in hitting the
targets on the computer screen with the cursor is a measure of the
degree of control each subject has over the beta rhythm. Subject
accuracy rates reached a plateau of . 90% after just six training
sessions.
Vol 10 No 8 3 June 1999 1769
NeuroReport
achieved an average accuracy rate of 94.2  4.2%
(s.d.), subject AB-1 achieved an accuracy rate of
95.0  4.3% and subject AB-2 achieved an accuracy
rate of 90.9  4.1%. Figure 2 shows that there was a
learning period involved for all subjects. This period
of time is restricted to the ®rst six sessions (,2±3
weeks). Control over the beta rhythm was main-
tained at a consistent accuracy rate for as long as 17
weeks in subject NP-1 and for at least 8 weeks in all
subjects.
Neuroprosthetic issues: Table 2 shows the effect of
movement upon the subjects' ability to control the
beta rhythm as measured in the accuracy rate. The
values given are the average accuracy rate from the
three runs in which that condition (non-movement,
right side movement, or left side movement) was
maintained. From these data it can be concluded
that extremity movement had little effect upon the
subjects' ability to control the EEG signal. For
subject AB-1, there was no effect of movement upon
accuracy rate. For the other subjects, there was a
change in the accuracy rate; however, the difference
between the averages was not statistically signi®cant
(Student's t-test, á ˆ 0.005).
The effects of neuroprosthesis operation upon
beta rhythm control were also limited. The overall
accuracy rate for subject NP-1 when the neuro-
prosthesis was active was 93.5  4.1%, which was
only 0.7% lower than the subject's average accuracy
rate without the neuroprosthesis (94.2  4.2%).
However, given the large standard deviations which
exist under both conditions, the effect of the neuro-
prosthesis upon beta rhythm control was not statis-
tically signi®cant (Student's t-test, á ˆ 0.005).
Neuroprosthetic operation: The neuroprosthesis
user was able to manipulate all three objects using
the EEG-based controller with his neuroprosthesis.
To achieve hand opening, the subject was instructed
to think of moving the computer cursor up, and to
think of moving the cursor down to make his hand
close. The transition between thinking cursor move-
ment and generating hand opening and closing
required ,10 min of training. However, the subject
did state that there were times where he would think
Table 2. The effects of upper extremity movement upon beta
rhythm control. As before, the degree of beta rhythm control is
measured by accuracy rate. For all subjects, there was little to
no effect of extremity movement upon beta rhythm control.
Subject Non-movement Right Left
AB-1 100% 100% 100%
AB-2 91.5% 94.8% 96.1%
NP-1 96.3% 92.1% 96.3%
1770 Vol 10 No 8 3 June 1999
R. T. Lauer, P. H. Peckham and K. L. Kilgore
of the wrong direction, and thus generate the
opposite command to the neuroprosthesis. Another
dif®culty encountered was the fact that the algo-
rithm used did not provide a means by which the
subject could lock his hand in the closed position
once the object was acquired. Therefore, the subject
had to continually think `hand closed' while manip-
ulating the object, which the subject stated was
increasingly dif®cult as the session went on (,1.5±
2 h testing).
Discussion
The results from this study demonstrate that the
EEG signal can be used to operate a neuroprosthesis
for the restoration of hand function. The fact that
the electrical stimulation of the muscles and remain-
ing voluntary arm movement did not interfere with
the quality of the control signal was important to
ascertain to validate further investigation in this area.
If either of these factors had interfered with the
quality of the EEG signal, then the feasibility of this
as an alternative control signal would be greatly
limited.
There are several questions which need to be
addressed if frontal beta is to be used to operate the
neuroprosthesis. First, since the signal is being
recorded from the frontal areas, there is the concern
that this is not `true' EEG control, but that the
subjects are using the electromyographic (EMG)
signal which can be generated by the frontalis
muscles to generate cursor control. This possibility
has been ruled out with these subjects since topo-
graphic and spectral analyses of the EEG signal did
not indicate the presence of muscle activity. Future
investigations will address this more fully. Second,
the command±control algorithm used to convert the
EEG signal into hand opening and closing only
provided dynamic hand function, i.e. the hand was
either open or closed. To achieve full bene®ts from
the neuroprosthesis, the user must be able to achieve
hand positions between either hand open or hand
closed, as well as have the ability to lock the hand in
place once a desired hand posture is achieved. These
issues will be addressed as the software and hard-
ware interfaces continue to evolve.
Finally, it is not known whether the EEG signal
will be able to provide multiple control signals to
allow for the operation of a bilateral neuroprosthesis
(i.e. a hand grasp system implemented in both
hands). Studies by Wolpaw [16] indicate that sub-
jects can learn to independently control the EEG
generated by each hemisphere. This would mean
that one hemisphere could control the operation of
one hand, independent of the other hemisphere.
However, these studies were conducted with the mu
EEG-based control of a hand grasp neuroprosthesis
rhythm recorded from the somatomotor cortex.
Whether this level of control can be achieved with
the frontal beta rhythms has yet to be investigated.
However, if this bilateral activation cannot be
achieved, there are still control algorithms which can
be implemented once the characteristics of the beta
rhythm are better de®ned which would allow the
subject to switch between the right and left hand
and use the same site to control hand function.
The results from this study provide information
that is not only useful for the ®eld of neuropros-
thetics, but also for assistive communication devices.
This study focused on training subjects to control the
beta rhythm component of the EEG recorded from
the frontal cortex to operate an external device (ini-
tially cursor movement, but later hand function via a
neuroprosthesis). The frontal beta rhythm was se-
lected because there should be little effect of extre-
mity movement upon the subjects control of the
signal. The previous work in this area by Wolpaw
and Pfurtscheller focused primarily upon using the
signal generated by the somatomotor cortical areas.
Although this is ideal for the target population in
their studies, signals recorded from this area cannot
be used for the operation of a neuroprosthesis. This
is because the mu and beta rhythms recorded from
these areas are naturally attenuated by extremity
movement [22,23], which would greatly interfere
with the subjects control over the signal. As demon-
strated in this study, the frontal beta rhythm is not
affected by movement. Also, subjects can be trained
just as easily to use the frontal beta as they can the
mu or central beta, and can achieve results which are
just as good as those in the other studies [13]. These
®ndings now provide evidence of alternative sites
which could be used instead of, or to augment, mu or
central beta rhythm control.
Speculation of the form of a cortical control inter-
face with a neuroprosthesis is premature, but some
elements are de®ned as a result of this work. First, the
recording areas, number of electrodes, and frequency
band for recordings are de®ned. In a viable implemen-
tation for an `out of laboratory' use, it is improbable
that the external recording array would be found
acceptable. However, it would be possible to imple-
ment such a recording array with surgically placed
subdermal electrodes and an external telemeter.
Conclusions
The analysis of the data indicates that the use of the
frontal beta rhythm is viable for operation of the
NeuroReport
neuroprosthesis. Subjects demonstrated that they
could achieve a high degree of accuracy with the
signal, and could maintain this level of accuracy
while generating voluntary movements or while the
neuroprosthesis is in operation. The answers to these
questions were critical in determining whether the
EEG signal was feasible as a control source for the
neuroprosthesis. The results from the ®nal study
demonstrate that the EEG signal is a feasible option
for controlling the neuroprosthesis. Further work is
underway to develop algorithms to convert the EEG
signal into a neuroprosthetic control signal which
would allow subjects to maintain their hold upon an
object for a long period of time, as well as to
provide ®ner control over the amount of hand
opening and closing (i.e. to achieve positions be-
tween fully open and fully closed).
References
1. Peckham PH, Poon CW, Ko WH et al. IEEE Trans Biomed Eng 28, 530±536
(1981).
2. Smith B, Peckham PH, Roscoe DD et al. IEEE Trans Biomed Eng 34, 499±508
(1987).
3. Keith MW, Peckham PH, Thrope GB et al. J Hand Surg 14A, 524±530 (1989).
4. Keith MW, Kilgore KL, Peckham PH et al. J Hand Surg 21A, 89±99 (1996).
5. Kilgore KL, Peckham PH, Keith MW et al. J Bone Joint Surg 79A, 533±541
(1997).
6. Johnson MW, Peckham PH, Bhadra N et al. IEEE Trans Rehab Eng in press.
7. Peckham PH, Mortimer JT and Marsolais EB. Ann Biomed Eng 8, 369±388
(1980).
8. Hoshimiya N, Naito A, Yajima M et al. IEEE Trans Biomed Eng 36, 754±760
(1989).
9. Handa Y, Handa T and Nakatsuchi Y. Jap J Med Elec Biol Eng 25, 292±298
(1985).
10. Kennedy PR, Mirra SS and Bakay RAE. Neurosci Lett 142, 89±94 (1992).
11. Kennedy PR and Bakay RAE. NeuroReport 9, 1707±1711 (1998).
12. Schwartz AB. J Neurophysiol 70, 28±36 (1992).
13. Humphrey DR and Tanji J. What features of voluntary motor control are encoded
in the neuronal discharge of different cortical areas. In: Humphrey DR and Fruend
H, eds. Motor Control: Concepts and Issues. New York: John Wiley and Sons,
1991: 413±444.
14. Chapin JK and Nicolelis MA. Electroencephalogr Clin Neurophysiol Suppl 45,
113±112 (1996).
15. Wolpaw JR, McFarland DJ, Neat GW et al. Electroencephalogr Clin Neurophysiol
78, 252±259 (1991).
16. Wolpaw JR and McFarland DJ. Electroencephalogr Clin Neurophysiol 90,
444±449 (1994).
17. Pfurtscheller G, Flotzinger D, Pregenzer et al. Med Prog Technol 21, 111±121
(1995±96).
18. Kuebler A, Kotchoubey B, Hinterberger T et al. Exp Brain Res 124, 223±232
(1999).
19. Kuebler A, Kotchoubey B, Salzmann H et al. Neurosci Lett 252, 171±174 (1998).
20. McFarland DJ, McCane L, David S et al. Electroencephalogr Clin Neurophysiol
103, 386±394 (1997).
21. McFarland DJ, McCane L and Wolpaw JR. IEEE Trans Rehab Eng 6, 7±11
(1998).
22. Kozelka J and Pedley T. J Clin Neurophysiol 7, 191±207 (1990).
23. Niedermeyer E. The normal EEG of the waking adult. In: Niedermeyer E and
daSilva, FL eds. Electroencephalography: Basic Principles, Clinical Applications
and Related Fields. Baltimore-Munich: Urban and Schwarzenberg, 1982: 71±91.
ACKNOWLEDGEMENTS: This study was funded by the NIH Neural Prosthesis
Program, the Ron Shapiro Charitable Foundation, and the Movement Disorder
Foundation. This work was conducted in the FES Center, a consortium of CWRU, the
Cleveland VAMC, and MetroHealth Medical Center. The authors acknowledge Dr
Jonathan Wolpaw and Dr Dennis McFarland for their assistance in the start-up of
these studies.
Received 17 February 1999;
accepted 13 April 1999
Vol 10 No 8 3 June 1999 1771