The Individualization of Training
by Siegfried Othmer | February 8th, 2007
The individualization of training has had a long history in neurofeedback. It began perhaps with Joel Lubar’s choosing whether to reinforce the standard SMR band of 12-15Hz or the low-beta band of 15-18 Hz, or whether to do both in pursuit of different objectives in work with ADD/ADHD children. At our hands, it eventually became a matter of choosing whether a person should train “SMR” at C4 or “beta” at C3, or perhaps a mixture of both. There seemed to be a general tendency toward lower-frequency training at the right hemisphere. We were able to address hemisphere-specific function in a more optimized way, and we could modulate arousal level more flexibly with the two available choices.
The two standard bands had both been given currency by Barry Sterman’s early work, and we took them as a given. The SMR band acquired a kind of usage validity over time, and the beta band could be considered a vernier. Margaret Ayers used the beta band almost exclusively in those early days, principally for working with head injury, and for her the SMR band was the vernier. The beta band was for nudging the under-aroused head-injured and ADD people (the Satterfield model) into better functionality, and the SMR band was for general calming of over-arousal.
Eventually we installed a vernier at 15 Hz to give us finer control over the modulation of arousal. That turned out to be so useful that the question became, why stop there? We then had to persuade our software writer on staff to take time out from his more ethereal exertions for the mundane task of giving us frequency selectivity at the 0.5 Hz level more or less across the board. This opened up the agenda for a wider range of arousal regulation, but we remained largely within the neighborhood of the standard bands.
Migration to lower EEG frequencies opened up the whole domain of affect regulation, particularly in the right hemisphere. At the same time, our clinical population was shifting from standard issue ADHD kids to those where behavioral inhibition and affect regulation were the dominant issues. Increasingly, we found ourselves training lower than standard reward frequencies. Behavioral volatility also drove us to adopt inter-hemispheric placements, as these were more stabilizing. Being stronger in their effect than our prior methods, they were also more frequency-sensitive. To get to stable brain states, we had to operate at the appropriate reward frequency. A coherent picture emerged for frequency rules that held for different homotopic site pairs.
Unsurprisingly, the new frequency rules for inter-hemispheric placements carried over adroitly to lateralized site pairs as well, in all of which we used T3 and T4 as a common site. One can just think of this in terms of the ear reference migrating up to T3 and T4. We now had an active site with a geometry not very different from the historical past. The old arousal model still held, in the sense that those whom we deemed over-aroused tended to train at the lower frequencies. But there was also a certain tautology in our thinking: anyone trained at very low frequency would be regarded as over-aroused.
In practice, the frequency range over which we trained kept increasing, and the arousal model had to be stretched to accommodate. Eventually we found ourselves training many people with the lowest filter bandwidth we had available, 0-3 Hz. (We had been operating standardly with 3-Hz wide filters for responsiveness.) For more than a year we were stuck there with quite a number of people for whom even this training was insufficiently calming. But we appeared to be up against the stops.
With our transition to BioExplorer software, we were now able to narrow the filter bandwidth on the fly in order to get to lower effective center frequencies. This allowed us to make immediate within-session gains with those folks with whom we were stuck at 0-3Hz. There seemed to be a whole new universe to explore at or below 1.5 Hz in center frequency. With the Neuroamp sporting a dc design, we were able to work effectively down to center frequencies as low as 0.05 Hz. This is an entirely different world from that of 12-15 Hz. And yet people were clearly still responding in a frequency-specific manner. Those who responded best at 0.05 Hz might well not like going back to 0.1 Hz. This could be repeatedly tested–blindly–to anyone’s satisfaction.
Recently Sue undertook to survey the 127 active clients we’ve seen in our clinic over the last six months to see how they sorted out in terms of reward frequency. The results are shown in Figure 1. The data fall roughly into two bins, those who train somewhere within the 0-3Hz domain and those who train at higher frequencies. And surprisingly, more than half of our current clients were ending up in the lower frequency bin. This is shown in Figure 2.
If we look at the two domains separately we observe patterns emerging that could guide both our models and our path forward. First with regard to the higher-frequency training, the data are shown in Figure 3. We observe a broad distribution with a peak around 7 Hz and perhaps another peak in the vicinity of 10 Hz. And when we survey the region of our prior standard bands we find only three out of the 127 clients. Only on the order of one to two percent of clients end up training optimally at the frequencies which we used to inflict on everyone. Another observation is that optimum frequencies above 12 Hz are a relative rarity.
The distribution for the low-frequency region is shown in Figure 4. Essentially half of all the data points fall into the lowest bin, with a reward center frequency of 0.05 Hz. This amounts to one-third of all the clients we have seen in the last six months. Certainly this qualifies as a major new departure in the design and execution of neurofeedback. If we put Figures 3 and 4 together we find that there is a no-man’s land at this point between 0.8Hz and 2.5 Hz. This could be a problem of poor statistics, or it could indicate that we are looking at two distinct mechanisms for the two frequency domains.
The story that cannot be told by the Figures is that the new training protocols have broken a number of barriers to progress among our clients, barriers posed by the prior 0-3 Hz limitation. Client satisfaction is greater overall; so is client commitment to the training. Boredom subsides as optimum reward frequencies are acquired. And clinician engagement is increased because there is more to work with and the details matter more. Training at the optimum frequency is clearly stronger than training elsewhere, and consequently there is more of a premium on actually localizing the optimum. Engineers in our midst will recognize that we are describing the characteristics of a hi-Q resonant system. Such a system can be excited at any frequency, but it is more more efficient to excite it at its resonant frequency. The excitation is not the point, however. The effect is, rather, to provoke a response by the brain to the mild challenge with which it has been presented. That response will be in the direction of improved self-regulation.
Neuronal assemblies must be organized in the frequency domain as resonant systems. And it is an empirical finding that in each individual one of these resonant systems has more clinical relevance than all the others. Exercising that system in turn renormalizes function across the entire frequency domain by virtue of the fact that all these regulatory mechanisms are coupled. What differentiates training at the optimum reward frequency is not efficacy per se but rather clinical efficiency. That is to say, none of the prior findings in neurofeedback using earlier techniques are hereby put in question.
The low-frequency training puts us in the spectral range of slow cortical potentials. It is appropriate to ask, therefore, whether we are converging with the Tuebingen group that has been using slow-cortical potential training all along. There remains an important difference. The Tuebingen group rewards for the transient achievement of a certain alteration in measured field potential. That very transient takes us out of the realm of slow cortical potential regulation! Rewards can only be achieved through the invocation of processes that have rapid response capability.
Only by working in the frequency domain as we are doing can we even speak of true slow-cortical potential training, because only this kind of training remains within that domain throughout the process. Ironically, then, we may not be the last to come to true SCP neurofeedback training, but rather closer to the first. But that does not really matter. The important points are 1) that our clinical reach has been vastly extended by including the low-frequency domain, and 2) that we remain thorougly tethered to frequency-based analysis in order to come to terms with the new findings.
