Feedback Electroencephalography and Sensory
Data Capture
Thomas B. Mulholland
Introduction
This chapter describes the history, applications and speculative
interpretation of one kind of feedback electroencephalography, an EEG
alpha-contingent stimulation system (ACS) which was developed for research
on an EEG component of the human orienting response - alpha blocking - (Mulholland
& Runnals, 1961; 1962b; Sokolov, 1963).
A posterior EEG alpha rhythm (8-13 cps) is recordable from 88-96% of adults,
in youth (a mature EEG is evident after age 12) and old age, in men and
women,and in all ethnic and racial groups. For a comprehensive review of the
normal EEG see Lairy (1976). A minority of humans (4-12%) have little or no
alpha frequencies in their EEG (Gibbs, et al, 1943; Vogel and Fujiya, 1969)
and may include a genetically anomalous sub-group (Anokhin, et al, 1992).
One of the familiar features of a normal, waking posterior EEG is a specific
response to sensory stimulation, the classical "alpha-blocking" response. If
posterior alpha rhythms (813 cps) are occurring and a sensory stimulus is
presented, especially a visual one, alpha frequencies are markedly
attenuated and an interval of low amplitude, mixed frequencies is observed.
For a serial sample of alphablocking responses, the latency (timedelay
between a stimulus ON and the start of alphablocking) is reasonably stable;
the timedurations of alphablocking are much more variable.
ACS is a biofeedback method by which stimuli (usually visual) which cause
alphablocking are made to be contingent upon the occurrence of EEG alpha
frequencies. It is however, an application of biofeedback method different
from the more familiar methods for training a person to control the spectrum
of their EEG (Kamiya, 1969). Whereas the latter emphasizes an EEG frequency
as an emitted response, an operant whose probability can be increased or
decreased by proper training, ACS emphasizes the suppression of alpha
frequencies following a sensory stimulus as an elicited response, a
respondent. A time series of alpha-blocking and alpha segments of an EEG is
less variable when it is obtained with ACS compared to other schedules of
stimulation and this means increased reliability of that dependent variable.
Interestingly, a feedback EEG experiment was reported early in the history
of EEG (Adrian & Matthews, 1934). See Mulholland (1969) for other early
examples of feedback EEG.
From the beginning of EEG science, alpha blocking was believed to be
associated with a psychological process of attention (Berger 1927). Alpha
was somehow linked to lower attention levels while alpha blocking or "low
amplitude, fast activity", was linked to higher levels of attention. The
"attention" hypothesis proved to be quite robust, seemingly immune to
evidence against it (Evans & Mulholland, 1968). Recently, in an incisive
review, Shaw (1992, 1994) has challenged traditional assumptions about the
origin and function of EEG alpha and alphablocking and their relation to
"attention".
At the same time that theories of alpha-blocking in relation to attention
were being proposed, new theories about integrative processes in the brain
and their relation to perception, attention, and consciousness were
developing in neurophysiology. For a recent theoretical discussion, see
Jasper, (1991). The question of the neural basis of the human EEG, from the
microstructure of cells to cell assemblies is still an active topic in
current research and theory. For reviews see Lopes da Silva, (1991) and
Steriade, et al, (1990).
Throughout the history of EEG science, studies of alpha and alphablocking
were often encumbered by assumptions which were either unproved, or ad hoc,
and which were hidden rather than explicitly stated. Many of these
assumptions are still extant:
Assumption 1: EEG is "background", "noise", "spontaneous", i.e., a
continuous random variable which is statistically stationary. These
conventional assumptions are ubiquitous in the literature on the average
evoked potential (Pivik, et al, 1993). Experiments have shown however, that
they are not defensible assumptions; for instance, the spectral mix of a
typical EEG is not statistically stationary for more than 815 seconds. (Praetorius,
et al, 1977; Basar, 1990). The occurrence of alpha in the posterior EEG is
not necessarily random; durations of sequential alpha and not-alpha segments
are not necessarily random; and alpha rhythms are not describable as "noise"
as will be shown in this chapter (Mulholland & Goodman, 1980).
Assumption 2: All alpha frequencies are equivalent. This is demonstrably
false. There are many alpha frequencies having different cortical origins
and recorded from different locations on the scalp which respond differently
to stimulation and motor acts (Cooper & MundyCastle, 1960; Walter, 1969,
1950; Lehmann, 1971, 1984; Kuhlman,1978; Steriade, et al, 1990; Gratton, et
al, 1992; Michel, et al, 1992). A lumping of alphas into one
undifferentiated class "the alpha rhythm" was often found in the early
biofeedback and psychophysiological literature. Alpha rhythms in this
chapter refer to occipital and occipital parietal alpha rhythms recorded
with socalled bipolar electrodes.
Assumption 3: Alpha-blocking can be interpreted as being caused by or
associated with increased "attention". The concept of attention is a
classical and still important concept in psychology and behavioral science.
Its referent is a subjective mental state or a class of behaviors, e.g.,
orienting responses. Acceptance of the term "attention" as an explanatory
concept was part of EEG science from the beginning (Evans & Mulholland
1969). The reality is however, that the term attention has so many meanings
it has minimal utility as an explanatory concept in feedback
electroencephalography. It is too easy to reify; too difficult to specify.
For this reason, the author has adopted a new term which is both neutral and
has fewer meanings "sensory-data capture" which, as will be shown later in
this Chapter, can be observed and measured. It can be conceptualized as a
sub-process which is associated with alpha-blocking in the posterior EEG and
which can be mobilized as part of a superordinate attention process.
Assumption 4: Alpha is associated with low arousal, low attention;
alpha-blocking is associated with higher arousal, activated attention. This
assumption is not necessarily true. Many researchers have shown that alpha
can be associated with higher and lower levels of arousal, vigilance, or
"attention" (Mulholland & Runnals, 1962a; Kreitman & Shaw, 1965; Mulholland,
1969; Creutzfeldt, 1969; Mulholland & Peper, 1971; Yingling, 1980; Collins,
et al, 1990)
Experimental evidence supports the hypothesis that posterior alphablocking
is primarily caused by an oculomotor control process, a mobilization of the
"triad of accommodation" (Mulholland & Peper 1971; Mulholland, 1969) and a
"retinal feedback component of oculomotor activity" (Wertheim 1974, 1981;
Shaw, 1992) which are subprocesses, part of a superordinate process of
sensorydata capture.
Assumption 5: The alphablocking segment which follows a stimulus is the
response, i.e., it is the singular response to a single sensory stimulus. An
alternative view however, better supported by experimental evidence, is that
an alphablocking response following a sensory stimulus is a disturbance of
an ongoing time series of alpha and alpha-blocking segments, a disturbance
which persists and decays after the stimulus is removed (Mulholland &
Runnals, 1964; Mulholland & Gascon, 1972). Bagchi, (1937) termed this
phenomenon "intermittent adaptation". Thus, one stimulus can produce an
alteration of time duration of each of a subsequent series of alpha-blocking
and alpha segments (Mulholland & Runnals, 1964).
Assumption 6: Within a specific alphablock, all parts of it, the beginning,
middle, and end are associated with the same or similar processes. This
assumption has been termed the "functional similarity" hypothesis (Mulholland,
1972). An alternative view which can be defended logically but which needs
to be examined experimentally, is that from beginning to end of a single
alphablock, many different processes occur, cascading and overlapping in
time, until alpha returns (Goodman & Mulholland, 1988). Spectral composition
however, may remain the same from beginning to end of an alpha-block (Goncharova
& Barlow, 1990) although this issue requires further research.
Assumption 7: Alpha segments (alternating with alphablocking segments) in a
time series are functionally equivalent with regard to the level of
alertness or arousal with which they are associated. Similarly, all
alphablocking segments are functionally equivalent. This assumption is
perhaps, an oversimplification. For instance, alpha at the beginning of a
timeseries following a stimulus may be associated with more alertness than
alpha blocking at the end, after several stimuli have been presented
(habituation) an hypothesis which needs to be examined further (Mulholland,
1972).
Assumption 8: Alpha and alphablocking are associated with specific opposite
psychological states. Certainly these are different EEG states, but it
remains to be proved that alpha-blocking is reliably associated with any
psychological states (Plotkin, 1976). Alpha is not necessarily associated
with an opposite state of decreased arousal or diminished vigilance as
compared to that which may be associated with alphablocking. Alpha seems to
be associated with a smaller set of states of cortical arousal compared to
alpha blocking, i.e., it is usually associated with a middling level of
cortical arousal. Thus alpha can occur when a sleepy person becomes more
wakeful; or when a vigilant person becomes less alert.
Assumption 9: Various sensory stimuli and different modalities, visual,
auditory, tactile, olfactory, etc., are equally effective with regard to
producing alpha blocking. It was recognized early in EEG studies that visual
stimuli were most effective for eliciting posterior alphablocking (Adrian,
1943). Auditory stimuli were less effective in this regard (Durup & Fessard,
1935). Also, vision is involved in the initial responses to other sensory
stimuli, e.g., we look toward the source of a sound, toward the location of
a touch, or to the origin of an odor as part of the initial exploration of a
new stimulus. If there is no visual component, the alphablocking response to
a sensory stimulus habituates quickly (although habituation to painful
stimuli is not proved). Alphablocking responses to visual stimuli however,
still occur even after many dozens of stimulus repetitions (Sokolov, 1963)
though the durations of each is less than earlier ones. Visual stimuli which
are signs or symbols can have a different effect on the alpha-blocking
process than stimuli qua stimuli (Mulholland & Davis, 1966; Sokolov, 1958).
In sum, research on the classical alphablocking response from 1925 1960
reflected a variety of assumptions and theories. Methods and procedures were
not standardized, definitions of psychological processes of attention were
inconsistent, and quantification of the response was neither detailed nor
specific enough. Researchers disagreed about the variables to be controlled
in an experiment and results from different experiments were not always
comparable. For this reason, many of the previous findings (or lack of them)
may need reevaluation with an improved method.
Alpha-contingent Stimulation: Method and Procedure
Before discussing the specifics of ACS method, there are general
methodological requirements for biofeedback and for EEG recording.
Biofeedback systems are sometimes wrongly described as if only a single
physiological process were part of the system. The term "alpha feedback" is
an example of an oversimplified, misleading designation. A biofeedback
system however, always includes a whole person, one who has an identity,
beliefs, unique expectations and perceptions, and idiosyncratic ways of
relating to and interacting with others. No one comes to biofeedback with
just their stated problem or goal. All of the personality traits,
prejudices, values and fears developed over a lifetime come along with their
EEG. It is a person who has a verbal and non-verbal dialogue with a
biofeedback practitioner, who interprets the meaning of the biofeedback
display and reacts to it and the setting in which biofeedback training
occurs. Without the recognition of a person in the system, biofeedback could
be an elaborate but superficial technology.
ACS method is used to study human data-capture which can be altered in
unpredictable ways by a participant's emotional reactions to the testing
laboratory and its staff. Staff must be carefully selected and trained so
that they can, by their skillful interactions with people, detect and
minimize anxiety and encourage a positive feeling or attitude. This permits
a volunteer to respond to the procedure and the enclosed, dark testing room
in a more predictable way with minimum contribution of fears, anxiety, etc.,
which can distort a typical EEG alphablocking response to stimuli or reduce
the amount of alpha too far below that volunteer's typical baseline. This is
especially important when children are examined.
Participants should also be given an opportunity to familiarize themselves
with the testing procedures and setting, e.g., they can come initially for a
"mock" trial. This can reduce anticipatory responses to the experimental
setting which can distort responses to ACS (Whittset, Robinson, & Kaplan,
1987).
The method of ACS requires meticulous adherence to the proper procedures of
clinical electroencephalography. This means that valid and reliable
recordings be obtained which are free of artifact and common kinds of
interference before the EEG is passed on to various filtering and signal
processing devices. The most common sources of artifacts and interference
are signals from the beating heart and contractions of the musculature,
movement of the conducting wires, from 60 cycle/sec power lines and from
interfaces of silver/silver chloride sensors, conducting gel and scalp.
Fortunately there are excellent texts available which describe common
artifacts and proper EEG technique (Niedermeyer & Lopes da Silva, 1987;
Pivik, et al, 1993).
Alpha contingent stimulation (ACS) was developed to standardize the method
for eliciting and quantifying the latency and duration of alphablocking
following sensory stimulation and, to control the response by reducing
unpredictable variation of the latency and duration of alphablocking. The
raw data are time durations of alpha and not-alpha segments in a time
series. Data analysis preserves the temporal sequence of EEG segments, i.e.,
the time-arrow, and in that way complements methods which pool sequential
samples of alpha and not-alpha segments (Pfurtscheller & Aranibar, 1977;
Pfurtscheller, et al, 1988). The method also facilitates the production of a
controlled coincidence of (1) a specific stimulus, (2) a cognitive process,
and (3) an EEG segment within a brief time frame.
The ACS method operates as follows. When alpha activity is above a preset
threshold of the amount of alpha frequencies in a participant's EEG it is
detected by a circuit which then switches on a stimulus. As described
before, the stimulus is followed by a blocking or attenuation of the alpha
rhythm for a variable time interval or duration. Alpha reoccurs and the
sequence is repeated as long as ACS is operating, resulting in consecutive
segments of alpha and "not-alpha", i.e., a time-series of EEG segments. See
Figure 1.
Figure 1 about here
The linking of EEG alpha segments to the occurrence of a stimulus, an
alphacontingent stimulus, was originally accomplished with a rather simple
configuration of a Grass Model III EEG machine, vacuum tube filters and
amplifiers, relays, and incandescent lamps, slide projector and a shutter
controlled by air-core solenoids (Mulholland & Runnals, 1961; Mulholland,
1962, 1969). EEGs were scored by hand by trained clerks who measured the
relay markers on the record to the nearest 0.1 sec. (Mulholland & Runnals,
1961, 1962b). The integration of a DEC PDP-12 computer with the ACS system
was accomplished in 1970. Software for collecting and measuring time series
of alpha and not-alpha durations was developed by Brad Cox (with 4K of
memory!) (Mulholland, 1962,1969). A next generation system was designed and
built by Boudrot (1972) which was integrated with a DEC PDP-12 computer and
an audiometric testing room. All of the software for that system (and
subsequent ones) was written and developed by Goodman (1973). A third
generation system was used after 1977, (Boudrot, Goodman & Mulholland,
1978). Today, an equivalent system can be configured using microprocessors
integrated with LSI circuit boards and off-the-shelf user-friendly software
for IO; signal detection, processing and analysis; data quantification;
creation of a display; and feedback control of the display. These
technological advances have made EEG biofeedback more accessible for
clinical applications in the fields of behavioral medicine and education.
The temporal relationships among an EEG, detectors of alpha frequencies, and
stimuli are schematically illustrated in Figure 2. t1 is the time delay
between recorded EEG alpha exceeding criteria for detection and the alpha
detector ON. In the studies discussed here, t1 was about 0.25 sec. t2, the
time delay between the alpha detector ON and the stimulus ON was usually
negligible. t3 is the time delay between stimulus ON and an EEG response
below criterion for alpha detection. This is a biological delay termed, in
the jargon of EEG, the latency of alpha blocking. The latency is usually
between 0.15 and 0.25 sec. (Cobb, 1963). t4 is the time between
alpha-blocking in the EEG and the alpha detector OFF. t5, the time delay
between the alpha detector OFF and the stimulus OFF, was negligible in the
studies presented here. t6 is the time delay between stimulus OFF and the
next alpha detector ON. This is termed the duration of alpha blocking, when
t5 is 0.
[Insert Figure 2 about here]
Whatever specific technology, is utilized, the EEG segment defined as
"alpha" must be in a specific frequency band (812 cps), have sufficient
power (>25 per cent of maximum recorded with eyes closed) and last long
enough (0.20.3 secs) so that an alpha segment is formed and detected and, a
stimulus can be presented. Stimulus ONOFF delays following the alpha
detector should be settable. Usually, the ON delay of the alpha detector
relative to onset of an EEG segment over threshold for alpha detection was
about 0.2 seconds. Recently, an improved alpha detector allows minimum ON
and OFF delays of the alpha detector of <0.1 sec. (Kodama & Mulholland,
1992). Experiments have shown that these values of filter frequency, rms
power threshold, alpha detector/stimulus delay so act that during feedback
stimulation, the variability of alpha and notalpha segments is minimized,
i.e., the system is properly "tuned" (Mulholland, Boudrot, & Davidson 1979).
Ideally, each participant's ACS system would be individually "tuned" so that
under ACS conditions, random variability of alpha and notalpha durations
would be lowest, i.e., replicative reliability would be highest. See Figure
3 for a flow diagram of a typical system.
[Insert Figure 3 about here]
Two different EEG-stimulus contingencies have been used in application of
ACS method termed Loop 1 and Loop 2:a configuration where a stimulus is
contingent upon alpha is called Loop 1; a reverse configuration where a
stimulus is contingent upon notalpha is called Loop 2. Usually, Loop 1 acts
like a timedelayed, quasi-negative feedback. Loop 2 performs like a
time-delayed, quasipositive feedback (Mulholland, 1968). Most of the
research reported here was done with Loop 1. A third kind of schedule of
stimulation which was non-contingent, was termed "sham". It was a schedule
of stimulation which was that obtained from a different participant under
ACS conditions or, a schedule which was created to have similar
interstimulus interval and occupy about the same total time, etc. A random
schedule was not used.
All experiments discussed here were done using an audiometric testing room
(8 X 12 ft) which minimized ambient visual and auditory stimulation.
Comfortable seating facilitated a relaxed posture and minimal movement.
Continuous monitoring of behavior and EEG of the volunteer ensured that
excessive movement, level of wakefulness or sleep and EEG artifacts could be
observed. Continuous monitoring can be accomplished by TV cameras in the
soundreduced room and with online, strip chart or videographic displays of
the EEG.
In most experiments, visual stimuli were produced by photographic
transparencies and a projector with an electronically controlled shutter.
Alpha detector ON causes stimulus ON; alpha detector OFF causes stimulus
OFF. Any type of visual stimulus can be presented or illuminated contingent
on the detection of alpha, e.g., words, text, photographic transparencies,
TV displays, computer graphics, and even human volunteers who, when
illuminated in the dark by an alpha-contingent light, become a social
stimulus (Mulholland, McLaughlin & Benson, 1979).
Sometimes, alphacontingent stimulation may not be a feedback. For instance,
the person viewing the contingent stimulus may change his/her way of
capturing sensory data, e.g., the volunteer may not look at the stimulus or
may try to capture data from other stimuli such as sounds or an audible
message. When this happens, a person viewing ACS may go "out of the loop".
Then the system is not a biofeedback system and becomes "dis-regulated"
(Schwartz, 1972; Mulholland, et al, 1983). The fact of error correcting
feedback can not be deduced from the fact of contingency; rather it must be
demonstrated by the kind of control which is achieved (Mulholland, 1977).
The procedure for evaluating the EEG during an experiment was standardized
for most applications. A participant's EEG was recorded with eyes closed
until alpha rhythms were clearly evident in EEG. The alpha detector's
threshold was set and the system was operated without presenting any
stimulus until about 10 alpha and 10 not alpha segments had been collected.
Then the participant was asked to open his/her eyes and the system continued
operating without presenting a stimulus until about 10 alpha and not alpha
segments were recorded. After these "eyes closed" and "eyes open" baselines,
ACS and "sham" feedback conditions were evaluated. In each of these
conditions 30 alpha and 30 not-alpha segments were recorded. Sham feedback
was a schedule of visual stimulation which was not contingent upon the alpha
rhythm. The stimulus durations and the interval between stimulations were
similar to those obtained with ACS stimulation.
An experimental trial or condition was defined by 10 alpha and necessarily
10 not- alpha segments with no stimulus and with eyes open in the dark
(Before feedback) continuing without an interruption into ACS or sham
feedback until 30 alpha and 30 not alpha segments were collected (During
feedback). An experimental session was made up of several trials. At the end
of the session, eyes opened and eyes closed baselines were recorded again.
When research on the ACS method began (Mulholland & Runnals, 1961) a
consensus of EEG research literature held that alpha rhythms were abundant
with eyes closed in the dark, that alpha rhythms were blocked initially when
eyes were opened in the dark, and that alpha-blocking followed a visual
stimulus though there were exceptions (Shaw, 1992). It seemed reasonable
then that ACS would be best done with eyes closed, i.e., stimulating through
closed eyelids. Subsequent experiments showed however, that copious alpha
could be recorded from volunteers whose eyes were open in the dark after a
few minutes (Mulholland 1965); moreover, alpha-blocking following ACS was
always followed by a return of alpha when the visual stimulus was removed (Sokolov,
1958).
With eyes open, visual stimuli which were especially effective for eliciting
data-capture responses, e.g., images of sexual or aggressive behavior and
meaningful verbal messages or words, could be presented. This permitted
studies of cognition and language not possible if eyes were closed. An
incidental benefit was also found with eyes open, participants didn't fall
asleep so readily! All subsequent studies after 1965 were done with eyes
open; eyes closed conditions were used primarily for baselines and
calibration.
The method of alphacontingent stimulation for evaluating sensory
data-capture has been used primarily with visual language and non-language
stimuli. EEG recordings are usually bilateral, most often from occipital and
occipitalparietal locations, O1, O2, O1P3, O2P4 in the standard nomenclature
for locating probes on the scalp (Jasper, 1958; Niedermeyer & Lopes da
Silva, 1987). Because alpha frequencies recorded from other locations on the
scalp are associated with different neurophysiological and behavioral
processes (Kuhlman, 1978), the results discussed here cannot be generalized
to other locations and refer only to occipital or occipitalparietal,
so-called bipolar EEGs unless indicated otherwise.
When these methods and procedures are used, valid, reliable recordings under
ACS conditions can be obtained from most children as young as 5 or 6 years
old (Mulholland & Gascon, 1972; Bryan, 1982) and from most adults up to 75
or more years old, and from patients who are diagnosed with neurological and
psychiatric illness (Mulholland, et al, 1976, 1979; McLaughlin, et al, 1974;
Bundzen, 1966; McLaughin & Koff, 1989). Obviously, ACS cannot be used with
those EEGs which contain little or no alpha but, as described before, these
are a minority of normal EEGs, from 4-12%.
Data Collection And Quantification
For the purposes of the ACS method, the waking EEG can be dichotomized as a
series of alpha segments alternating with notalpha segments. The series of
alpha and not-alpha segments are numbered beginning with the onset of an
experimental condition. Thus N1a is the first alpha segment detected; N2a
the second, etc. Similarly, N1na is the first not-alpha segment detected;
N2na is the second, etc. In most of the studies presented here, N during ACS
was 30 for both alpha and for not-alpha segments preceded by 10 alpha and 10
not alpha segments collected without ACS. These "before" and "during
feedback" samples of EEG segments defined a trial.
It can be seen in Figure 2 that the duration of an EEG alpha segment is
approximately equal to t1 ON delay + t3 (latency) - t4 OFF delay and that
the duration of EEG alpha blocking is approximately equal to t4 OFF delay +
t6 - t1 ON delay, when t2 and t5 are negligible. Time durations of alpha
detector ON and OFF were good approximations to durations of EEG alpha
segments and of not-alpha segments because alpha detector ON and OFF delays
are approximately equal.
The successive time durations of alpha and not-alpha segments were collected
and displayed on-line as computer graphics as two separate samples. Usually,
this was done for both the so-called connected EEG and the contralateral
homologous EEG which was not connected to the feedback system.
As described previously, alpha and not-alpha segments occur alternately in a
trial. Bestfit functions computed in each trial were used to estimate the
trend of alpha durations and separately, of notalpha durations as a function
of the ordinal number of the EEG segment both before and during feedback.
These functions, described below, were empirically selected and were not
derived from any particular theory. Other polynomials were fitted and some
fitted equally well; without a theoretical justification the choice of a
particular polynomial is arbitrary. Estimates of trends from each trial are
pooled to obtain average bestfit functions of latency and duration of
alphablocking. In the computer graphics and in some of the illustrations in
this chapter, the trends were plotted as a continuous function; however,
values of the function exist only for the ordinal numbers of the segments in
a series.