Psychophysics of EEG State Discrimination
(Draft manuscript on research in progress, 4/3/2006)
Introduction
Perceptual discrimination is the ability to accurately report objective differences between stimuli based on subjective perceptions. The standard paradigm is in EEG learning theory is operant conditioning to teach people to control EEG states. In this paradigm, brainwave information is displayed in real time on a monitor and subjects are rewarded when some value, usually power or amplitude, exceeds a threshold set by the trainer. Discrimination training is also a form of operant conditioning, where the conditioned response is perceptual rather than behavioral.
One of the earliest and often cited studies in EEG learning was a discrimination experiment by Joe Kamiya (1970), who reported success in training human subjects to discriminate alpha from non alpha states. In this study, subjects were asked to respond "A" for alpha and "B" for non alpha when the experimenter rang a bell. The experimenter waited for distinct alpha and non-alpha states, in random order, to appear in the raw EEG and then rang the bell when they appeared. Kamiya reported that 9 out of 12 subjects reached a significant proportion of correct within seven one-hour sessions.
Interestingly, nearly all research on human learning of brainwave states since this study have focused on training and measuring voluntary control of EEG constructs, rather than discrimation. Psychophysiologists have, however, measured discrimination learning outside the central nervous system. Discrimination has been demonstrated for finger temperature (Lombardo & Violoni, 1994), heart rate (Grigg and Ashton, 1986), cephalic vasomotor activity (Fudge & Adams, 1985), and blood Pressure (Martin & Epstein, 1980).
It is often stated that in the course of neurofeedback training, people increase their perceptual acuity for subtle internal signals about their current EEG state. However, it is clear that controlling and having awareness of a behavior are two separate, but intimately connected, processes in the brain. For instance, a professional musician could play a piece of classical music flawlessly, and yet still could not tell you the coordinated sequence of muscle contractions it takes to produce the behavior. In fact, if a musician or an athlete pays too much attention to the details of their performance, they fall out of "the zone" and are more likely to make mistakes.
The ability to attend explicitly to the details of one's sensations and behavior is clearly a limited resource in the brain. Researchers have observed that people can hold about "seven plus or minus two" unrelated items in short term memory at any given time. Attention and short-term memory tend to be allocated most to the early stages of learning. Once a behavior becomes well-established, its performance is delegated to subconscious systems of the brain.
So, does operant conditioning to train control over EEG states also teach awareness of these states? It is likely that some degree of awareness is achieved during most forms of control learning. Cinciripini (1984) demonstrated that subjects trained in 12-25 Hz control showed significant discrimination learning effects over no feedback controls. Meanwhile, Joe Kamiya's study showed that alpha discrimination training improved people's performance on alpha control tasks. So there is at least preliminary evidence that there is some interaction between discriminative awareness and voluntary control of EEG states. However, given the frequent assertion by biofeedback therapists that control training increases awareness, it is remarkable how few EEG biofeedback studies have actually measured whether trainees can correctly identify their internal state. (However, Wytze van der Zwaag recently emailed me and said, "Boris Kotchoubey has done (2002) a nice study with SCP on this topic.")
One study (Cott, Pavlski, and Black, 1981) failed to demonstrate discriminative learning of the alpha rhythm, but differed substantially from Kamiya's original study in defining an alpha state as one-half second of high absolute alpha power. Kamiya (personal communication) believes that one-half second is barely at the threshold of a subject's ability to discriminate a state, and that absolute EEG measures (unlike relative measures or visual inspection of the raw EEG) allow for the contamination of the spectral content of the "alpha" band. Therefore, it remains to be seen whether discrimination can be trained using relative alpha and a longer discriminative stimulus interval such as two seconds.
Rhythmic auditory and visual stimulation has been demonstrated to evoke changes in the EEG corresponding to the frequency of stimulation (see Frederick et al., 1999). Harold Russell (personal communication)suggests that this "entrainment" effect could serve to amplify the discriminative stimulus. Thus, we hypothesize that participants who are provided with fixed frequency AVS at the their dominant alpha frequency will achieve significant performance sooner and a higher level of performance on their tenth session (compared to those who do not receive AVS).
Method
In a preliminary study of one subject, EEG was recorded at Fz with a linked ears reference using custom software and a Brainmaster EEG. Two hundred second baselines were recorded, followed by the task condition. EEG relative amplitude in the theta (4-8 Hz), beta (14-21 Hz) and alpha (11-14 Hz) frequency bands was calculated for each epoch. The nonstandard frequency range for alpha was used because this subject's peak alpha frequency was 13 Hz. Percentile scores were determined by the ranking of a moving average of scores initially derived from the baseline. For the "theta/beta" task, a random order of high and low trials was generated, and a tone was sounded, requiring a response of "high" or "low" (pressing the H or L key on a laptop computer), whenever the relative amplitude exceeded a twenty percentile difference from the median of the baseline. In the "alpha" task, the subject was allowed to decide when to respond, pressing a "high" or "low" key whenever he felt a high or low alpha state was present. At least ten seconds was required between responses in the alpha task, and the subject had to alternate between high and low responses. Sessions were of 10 to 100 trials in length. A correct response was indicated if a "low" response was recorded when the amplitude was at or below the 50th percentile, or if a "high" response was recorded when above. The subject received feedback about whether the response was correct or incorrect after each trial.
Results
The success rate for the theta/beta task was insignificant, 53% correct out of 487 trials across fifteen sessions. The success rate for the alpha task was 66% over 1168 trials (binomial p<.0001). When histograms of relative alpha amplitude (percentile rank) vs. success were constructed in ten percentile bins, the perceptual threshold for alpha discrimination appeared to be 30%. That is, events in the 0-40 and 70-100 percentile amplitude events were guessed correctly (63-84%, p<.001), while events in between were guessed at chance levels. "High" alpha responses were more often correct (76%, p<.0001) than "low" (55%, p<.01). Average performance declined significantly across sessions during a given day, suggesting a fatigue effect. Across both tasks and all sessions, the correlation between session success and the number of previous trials that day was -.44 (p<.01).
Discussion
These preliminary findings replicate Kamiya's finding that alpha discrimination is a learnable response, and suggest that theta/beta discrimination might be learnable if naive subjects are allowed to decide when to respond rather than being randomly prompted. Future studies will characterize differences in perceptual acuity between subjects and among various EEG constructs and scalp locations. Specifically, the next phase of the study involves training four subjects in alpha discrimination at Pz, two subjects subjects in frontal alpha asymmetry discrimination, and two subjects in discrimination of frontal full-spectrum power (the "In-All" protocol). Two of the alpha discrimination subjects will receive photic stimulation at their dominant alpha rhythm. We hypothesize that photic stimulation will amplify the discriminative stimulus, resulting in a greater rate and extent of learning in these subjects.
The preliminary study differed from that of Kamiya (1970) in two important respects. First, alpha was defined in terms of relative spectral amplitude, whereas Kamiya visually detected alpha events in the raw EEG. Since Kamiya's method is rarely used in modern biofeedback practice, however, using relative amplitude may have greater clinical validity. Secondly, the subject chose when to respond in this alpha task, whereas Kamiya's subjects were randomly prompted to respond. The subject-initiated responding in this study confounds discriminative awareness with voluntary control, in that the subject can willfully evoke alpha/non-alpha states before responding. Subject-initiated responding was used for the alpha experiment because the experiment was performed after the failed theta/beta experiment. It was correctly assumed that subject-initiated responding would result in better performance. Further, this method still accurately measured whether the subject was correct in his perception of his volitional state and its results. Kamiya's protocol also did not preclude subjects responding correctly to random prompts because they had willfully evoked or suppressed an alpha state. As in the learning and performance of more ordinary behavioral skills, volition and awareness of EEG states appear to be separate, but inextricably linked processes. Studying the differences between random vs. subject-initiated prompting could help to illuminate the nature of this interaction. One goal of the present study is to assess whether skills learned during subject-initiated responding generalize to random prompt training.
One possible advantage of the interaction between the two forms of learning is that, insofar as discrimination is confounded with control, discrimination could be a more precise experimental model of control. Learning is difficult to measure in EEG control training because therapists often adjust thresholds to maintain an optimum percentage of reward, and effects of training are often smaller than the baseline variation. By contrast, a direct measurement of success is intrinsic to every trial in discrimination learning. Thus, if control and discrimination skills generalize to each other, taking EEG state discrimination measurements could be a useful method of assessing client progress for neurotherapists who train control, and discrimination training could potentially improve the extent and rate of learning in control training.
Training discrimination could also have therapeutic value in its own right, just as insight-oriented psychotherapy can have value above and beyond behavior-modification psychotherapy. It is likely that physiological states which can be controlled, and those which can be discriminated, are separate but overlapping domains. Knowing what state one is in might in some circumstances be of equal or greater importance than being able to control that state. For instance, being more aware of physiological states that suggest the imminence of, for instance, a manic or depressive episode, or an epileptic seizure, could help patients to respond appropriately in ways that do not involve changing the state.
We are currently accepting applicants to participate as experimental subjects. Here is the text of the subject recruitment flyer we are using:
----------------------
Can You Tell What State You Are In?
Research participants are sought for a study of human ability to discriminate differences in electroencephalographic (EEG) states. While EEG is being recorded, participants will be prompted to respond "high" if they are in a high amplitude state or "low" if they are in a low amplitude state, and then informed whether their responses are correct or incorrect. Previous studies have suggested that most people can learn to correctly discriminate EEG states at above chance levels within a few sessions. Participants are paid $10 per one-hour session for up to ten sessions. For further information, contact Jon Frederick, Ph.D., at [ smiile-at-runbox.com ].
-----------------------
Since funding is limited, I am hoping to find mature and dedicated subjects, and I hope that at least a few will have extensive training in mediation, athletics, or martial arts. Having one or two participants with prior neurofeedback control training could also provide a useful variation.
Generous donors-- or researchers who want a study of the psychophysical properties of their own favorite EEG constructs-- can sponsor one participant through all ten sessions for only $250. Donations are accepted by a 501-c3 nonprofit and are tax-deductible. Free software available for do-it-yourself-ers.