Академический Документы
Профессиональный Документы
Культура Документы
1, MARCH 1998
I. INTRODUCTION
(a)
(b)
(c)
Fig. 1. The three trial formats: standard (STD), no cursor movement (NCM),
and no feedback (NFB). The sequence of events during an STD trial is 1) a
target appears on the screen, 2) the cursor appears and begins to move, 3)
the cursor moves toward or away from the target, depending on the subjects
EEG, 4) if the cursor hits the target, the target flashes, but if the cursor misses
the target the target disappears and the cursor remains on the screen, 5) there
is a brief interval between trials, and 6) another trial begins as in 1). Thus, in
STD trials, feedback is present as both cursor movement and end result (i.e.,
a target flash for a hit or the cursor frozen on the screen for a miss). In NCM
runs, cursor movement feedback is absent. In NFB runs, there is no feedback
at all, neither the cursor movement nor the end result is shown.
TABLE I
EACH SUBJECTS ELECTRODE DERIVATION, CENTER FREQUENCY OF EEG CONTROL SIGNAL, ACCURACIES FOR SESSIONS 9 AND 10 AND FOR STANDARD (STD) AND
NO-CURSOR-MOVEMENT (NCM) RUNS OF SESSIONS 11 AND 12, AND r 2 VALUES FOR STD, NCM, AND NO-FEEDBACK (NFB) RUNS OF SESSIONS 11 AND 12
only the target and the terminal event signaling a hit or miss.
In two NFB (no feedback) runs, neither the cursor nor the
terminal event appeared on the screen (i.e., Fig. 1, bottom).
The subject saw only the target, which simply remained on
the screen for 5 s and then disappeared. No time limit was
imposed in the other conditions.
We compared all STD and NCM runs in accuracy (i.e.,
percent of targets hit) and the STD, NCM, and NFB runs in
value, the coefficient of determination [25], which indicates
the proportion of the total variance of the EEG amplitudes for
top and bottom targets that is accounted for by target position
(i.e., by the difference between the mean amplitudes for top
and bottom targets).
III. RESULTS
A. Initial Acquisition of EEG Control
As Table I shows, the ten subjects varied widely in the level
of EEG control they acquired over the first 10 sessions. Eight
of the 10 achieved some measure of control: their accuracy
for sessions 9 and 10 was significantly (
,
in all cases) greater than 50%. These results are
consistent with our previous experience, which indicates that
high accuracy (i.e., 90%) usually takes several months to
develop and that subjects vary greatly in their learning rates.
Table I also indicates that, for this relatively small number of
subjects, neither the EEG channel derivation used (i.e., bipolar,
CAR, or Laplacian) nor the presence of spinal cord injury
clearly affected success.
10
(a)
(b)
Fig. 2. (a) Mean ( SE) accuracies (in %) for the STD and NCM runs of
sessions 11 and 12, and (b) R2 values ( SE) for the STD, NCM, and NFB
runs of sessions 11 and 12.
that allows the subject to acquire control over the EEG signal,
and thus over cursor movement.
At the same time, however, feedback can have other effects,
good or bad, on performance. The multiple possible effects
of feedback have been addressed in a variety of studies [1],
[17]. Some effects are lasting, which is to say that they affect
learning. Others are transient, in that they only affect performance. It is not clear whether feedback is needed or desirable
in all phases of performance. Salmoni et al. [17] suggests that
excessive guidance can sometimes actually degrade learning.
It does seem clear that feedback should be as rapid as possible.
Delayed feedback degrades performance. Smith and Smith
[19] suggest that ideal humanmachine systems should provide
instantaneous feedback. This theoretical ideal is not possible
for EEG control, since frequency analysis requires some finite
sample of EEG data (e.g., 200 ms) and an additional 20 ms
is needed for computation and screen display. Our experience
to date suggests that this delay is not a major impediment to
performance. It is comparable to the delay inherent in most
humanmachine interfaces [2]. We have evaluated the effects
of signal length on alternative signal processing algorithms
with this issue in mind [10].
In the present study, short-term removal of feedback did not
have a significant effect on the overall performance of the subject population. Nevertheless, some subjects were affected by
removal of feedback, since performance was no longer significantly correlated with performance during previous sessions in
REFERENCES
[1] A. H. Black, The operant conditioning of the electrical activity of the
brain as a method for controlling neural and mental processes, in The
Psychophysiology of Thinking, F. J. McGuigan and R. A. Schoonover,
Eds. New York: Academic, 1973, pp. 3568.
[2] D. R. Bradley and S. B. Abelson, Desktop flight simulators: Simulation
fidelity and pilot performance, Behavioral Res. Methods, Instrum.
Comput., vol. 27, pp. 152159, 1995.
[3] L. A. Farwell and E. Donchin, Talking off the top of your head:
Toward a mental prosthesis utilizing event-related brain potentials,
Electroencephal. Clin. Neurophysiol., vol. 70, pp. 510523, 1988.
[4] H. Gastaut, Etude electrocorticographique de la reactivite des rythmes
rolandiques, Revenue Neurol., vol. 87, pp. 176182, 1952.
[5] H. H. Jasper, The tentwenty electrode system of the International Federation, Electroencephal. Clinical Neurophysiol., vol. 10, pp. 371375,
1958.
[6] J. Kalcher, D. Flotzinger, C. Neuper, S. Golly, and G. Pfurtscheller,
Graz braincomputer interface IIToward communication between
humans and computers based on online classification of three different
EEG patterns, Med. Biol. Eng. Comput., vol. 34, pp. 382388, 1996.
[7] J. W. Kozelka and T. A. Pedley, Beta and mu rhythms, J. Clin.
Neurophysiol., vol. 7, pp. 191207, 1990.
[8] W. N. Kuhlman, EEG feedback training: Enhancement of somatosensory cortical activity, Electroencephal. Clin. Neurophysiol., vol. 45,
pp. 290294, 1978.
[9] S. L. Marple, Digital Spectral Analysis With Applications. Englewood
Cliffs, NJ: Prentice-Hall, 1987.
[10] D. J. McFarland, T. Lefkowicz, and J. Wolpaw, Design and operation
of an EEG-based braincomputer interface (BCI) with digital signal
processing technology, Behavioral Res. Methods, Instrum. Comput.,
vol. 29, pp. 337345, 1997.
[11] D. J. McFarland, G. W. Neat, R. F. Read, and J. R. Wolpaw, An
EEG-based method for graded cursor control, Psychobiol., vol. 21, pp.
7781, 1993.
[12] G. R. McMillan, G. L. Calhoun, M. S. Middendorf, V. H. Schnurer,
D. F. Ingle, and V. T. Nasman Direct brain interface utilizing selfregulation of steady-state visual evoked response (SSVER), in Proc.
RESNA95 Annu. Conf., 1995, vol. 15, pp. 693695.
[13] G. Pfurtscheller and A. Berghold, Patterns of cortical activation during
planning of voluntary movement, Electroencephal. Clin. Neurophysiol.,
vol. 72, pp. 250258, 1989.
11
Dennis J. McFarland received the B.S. and Ph.D.degrees from the University
of Kentucky, Lexington, in 1971 and 1978, respectively.
He is currently a Research Scientist at the Wadsworth Center for Laboratories and Research at the New York State Department of Health, Albany, NY.
His major research interests are in development of EEG-based communication
and in auditory psychophysics.