What's New in Neurofeedback

• 10th Annual Winter Brain -Future Health

Yesteday as I reviewed a handful of recent QEEG publications, I was struck by the variance not only in methods (that's a given) but in channel displays, in communicating the basic electrode array. All of these papers included analysis of 19 channels of data, and the investigators employed the standard 10-20 system of electrode placement, but from there on in showing us the data from each channel seemed like dealer's choice. It reminded me of our own cultural Tower of Babel, in written form: some tribes write sentences left to right (e.g., Romance languages), some right to left (e.g., Hebrew), and some top to bottom (e.g., Chinese). Each writing style is a solution, an invention for transforming a linear, temporal dimension into a spatial one, a surface area, a page of text. Each solution has advantages and drawbacks, but at least readers of a language are not surprised by the basic direction or orientation of sentences even time they open a book.

    Three papers in the Journal of Neurotherapy: three different channel arrays. Can't we adopt a standard, like an APA format for EEG data -- at least for the journal if not the entire field? (Standards need not be followed when not following them adds information.) On the PC screen, freedom to move channels up and down, screw with orders, is a given, but when you transfer electrons to atoms (ink), the added permanence requires that we communicate as much as possible, across time and individuals. The development of a useful standard could aid in this goal.

  
         Fp1   Fp2
    F7  F3  Fz  F4  F8
    T3  C3  Cz  C4  T4
    T5  P3  Pz  P4  T6
         O1    O2

Pictured above are the 10-20 electrode positions for 19 active recording sites. Now, to tell the reader something about each site, we either color code the sites and interpolate between points, or if the information cannot be so easily conveyed, we transform this 2-D display into a 1-D array, a single column, and leave the remaining surface area for squiggles (raw data), spectral means, correlation coefficients, what have you.

Spatial proximity of the electrodes should be the starting point for any standard display. Placing Fz next to O1, for instance, should be avoided. What is gained from juxtaposing the behavior from widely separated and presumably functional disparate sites? Functional considerations may occasionally override spatial proximity, but given the heterogeneous methods of judging functional linkage between sites, promixity in space, not function, should constrain the array's order.

That said, we are dismantling a 2-D surface and drawing it out like taffy into a single dimension so some spatial proximity will have to be sacrificed as others are spared. F3 cannot be next to Fp1, F7, Fz, and Cz in a single line.

Here is a common array, what I call sagittal series.

F7
T3
T5
Fp1
F3
C3
P3
O1
Fz
Cz
Pz
Fp2
F4
C4
P4
O2
F8
T4
T6

Saguttal series is the ginsu knife approach of EEG science -- cut, cut, cut -- down the head's length, regardless of functional or evolutionary considerations. Such a system makes homologous site comparisons nearly impossible (e.g., F7 vs F8?). Excel could make more sense of brain activity by simply sorting sites alphabetically. At least this would keep brain regions together.

Another array found in JNT:

F1
F2
F7
F8
F3
F4
T3
T4
C3
C4
T5
T6
P3
P4
O1
O2
Fz
Cz
Pz

Pairing homologous sites provides a local elegance, with some global chaos. I would expect functional similarity between left and right homologs to be greater than between some adjacent sites (e.g., P3 and C3), but what is to be done with mid-sagittal sites, and how to order the pairs (the global chaose I mentioned)? And aligning the squiggles of Fz, Cz, and Pz together for comparison rarely provides insight into functional relevance and commonality as much as aligning Pz with its parietal neighbors would, for instance. I don't think this is the best array, but I did find empirical justification in magnitude values.

In order to investigate how magnitude values might reveal functional commonalities between sites, I first had to deal with a artifact inherent in magnitude estimation -- namely, reference effects. Spectral values are a function of the difference between an active electrode and its presumed inactive reference. If the reference were truly inactive then there would be no problem, but because there is activity in those silent sites, some shared activity in fact, distance between active and referential electrodes is very important. We see less difference (low values) when the active site is near the reference, and high values when it is far away. Ratios are a simple way to handle reference issues (while losing other factors), and a ratio of ratios makes it even more robust, or so I hope. The measure I decided upon for evaluating functional commonalities between sites is called task synchrony, for lack of a better name. Task synchrony refers to the relative difference in magnitude values between two sites across tasks. For instance, if 10.0 uV are recorded at F3 during eyes closed rest and 10.5 uV at F4, the difference is 5% (10.0-10.5)/10.0. Then during eyes open, F3 now presents 5.0 uV and F4 5.2 uV, the difference is now only 4%. The difference of the differences is 5% - 4% which equals only 1% variance between tasks. This implies a high task synchrony for F3 and F4, for the tasks evaluated. The similarity of spectral values between, say, F3 and T3, may be high during eyes open, due to referential and not functional issues, but low during eyes closed, which would produce a large difference and thus low task synchrony. This is just another measure of functional coupling, perhaps the least useful one to date, but it did spit out multiple homologous sites across the entire head without hesitation:

 Task synchrony for single wide band 1-24 Hz
   SITE  Best match  % differences |EC-EO|
    Fp1      Fp2     1.1
    Fp2      Fp1     1.0
    F7       F8      0.3
    F3       F4      0.0
    FZ       F4      0.2
    F4       F3      0.0
    F8       F7      0.3
    T3       T4      0.7
    C3       C4      0.3
    CZ       C4      3.4
    C4       C3      0.3
    T4       T3      0.7
    T5       P3      0.0
    P3       T5      0.0
    PZ       P4      1.0
    P4       T6      0.8
    T6       P4      1.1
    O1       T6      2.2
    O2       O1      3.4

Electrode site T6 shows unusual linkage to P4 and O1, and mid-sagital sites (the Z's) most closely resemble right medial sites, but what's most notably is the numerous homologous sites linking bidirectionally to each other.

Comodulation values across the spectrum (1-24 Hz) in eyes closed (n=50) and eyes open (n=34) baselines reveal a different pattern of functional similarity:

  GREATEST COMOD VALUE BY SITE (1-24 Hz)  

SITE   Highest Comod. Value (and which Site)
      __EC____   __EO____
Fp1   Fp2 0.84    Fp2 0.82
Fp2   Fp1 0.84    Fp1 0.82
F7    F3  0.75    F3  0.71
F3    Fz  0.88    FZ  0.86
FZ    F3  0.88    F3  0.86  
F4    Fz  0.88    FZ  0.84
F8    F4  0.74    F4  0.68
T3    C3  0.66    C3  0.61
C3    Cz  0.77    CZ  0.76
CZ    C3  0.77    C3  0.76
C4    Cz  0.77    CZ  0.74
T4    C4  0.65    C4  0.56
T5    P3  0.76    P3  0.73
P3    Pz  0.82    PZ  0.82
PZ    P3  0.82    P3  0.82
P4    Pz  0.80    PZ  0.80
T6    P4  0.73    P4  0.70
O1    T5  0.73    O2  0.72 
O2    P4  0.74    O1  0.72 

If we choose our channel display array based on spatial proximity and use functional relatedness as a tie breaker, we should be able to generate a display with optimal communicative and informational value. Let's do just that.

Using Comodulation as our functional index:
• Fp1 is almost equally distant from F3, Fz, and Fp2, but in terms of function it most resembles Fp2 during both eyes closed and eyes open rest (as the table above shows), so Fp1 should be next to Fp2 in our array.
• F7 is proximal to F3 and T3, but it is most alike functionally to F3 and thus F7 should be next to F3 in our display.
• F3 is most alike to Fz despite being proximal to many sites (Fp1, F7, Fz, and C3) so F3 should be next to Fz.

Fp1:Fp2 Fp2:Fp1 F7:F3 F3:Fz Fz:F3 F4:Fz F8:F4 T3:C3 C3:Cz Cz:C3 C4:Cz T4:C4 T5:P3 P3:Pz Pz:P3 P4:Pz T6:P4 O1:O2 O2:O1

One of the best solutions for fitting this pairing information together into a single line is already shown above, as the first column in the table. This array handles the issues of geometry and function reasonably well.

Now that I've trotted out comodulation, many will ask how does comodulation differ from coherence. Comodulation and coherence are two measures of functional relationships between sites (e.g., coupling or dependence), and they often present the same results, but they occasionally reveal differences. One difference from my perspective is that calculation of comodulation is easier to explain without resorting to Greek (like capital Sigma). Comodulation is calculated in the time domain; it is simply linear correlation between two phenomena. We call it comodulation because the phenomena we evaluate waxes and wanes, and thus the linear correlation estimates how much of this modulation is shared by sites. But one could theoretically measure comodulation of any two phenomena, such as the number of people born to number of human hands on planet (a comod value near +1.0), the comodulation of IQ and SAT scores, etc. Comodulation calculation does not require cyclical data, though it probably wouldn't be called comodulation then (instead it would be the correlational coefficient). Coherence calculation requires frequency data.

Nunez characterizes coherence as phase consistency between sites. Comodulation is magnitude consistency between sites. Often these measures behave similarly, but here is an example when they don't:

Channel 1:  16 04 16 04 16 04 16  uV
Channel 2:  04 16 04 16 04 16 04  uV

The waveform in one channel is exactly the same as the waveform in the other, except for amplitude. Thus coherence is +1.0 for these channels (and phase 0 degrees). Comodulation is not +1.0 but instead -1.0 as the magnitude in one channel perfectly negatively correlates with magnitude in the other channel. Viva la difference!

In Nunez and colleague's 1997-1999 opus, EEG coherency (Part I & II), they conclude that "studies of coherence and brain state should include several different kinds of estimates to take full advantage of information in recorded signals." Not unlike Jay Gunkelman's succinct mantra, "remontage, remontage, remontage" to get a clearer picture of the electrical activity under the skull. If remontaging provides non-trivial differences in information about brain activity, not negligible residuals, but large enough to build careers on, we are in for a long struggle in using EEG to assess brain function. There will always be some unremovable noise that derives solely from our referential technique.

Perhaps the best EEG measure is to combine all analyses into a single composite and quantify how well this composite predicts any part of the data set. If any small segment could well predict all aspects of a latter segment, we would have high predictability; the system would exhibit low spatiotemporal complexity. That, I suspect, cannot be good for the brain. In fact poor predictability of the whole from the part might be a good definition for complexity. The best measure of brain health would be unpredictability, or true randomness from moment to moment. The healthier the head, the poorer the predictability. (Of course at some levels of explanation, predictability would be helpful, such as circadian rhythms, etc).

Think of EEG as an irrational number extending across time. Any predictable (or rational) segments may indicate a breakdown of control or communication within the brain, much like a heart that normally exhibits a chaotic temporal function in its beatings when healthy but becomes achaotic immediately before an attack. As one electrophysiological system works, so might another.

So the take-home point of this brief article may be: chaos is good for the brain, but not for the mind.

News & Reviews NEW BOOKS

Stimulant Drugs and ADHD: Basic and Clinical Neuroscience
by Mary V. Solanto

Describes neuroanatomy and functional neurophysiology involved in dopamine and norepinephrine systems with respect to regulation of arousal, activity, and impulse control. -www.amazon.com/exec/obidos/ASIN/0195133714/top100

When the Brain Dies First
by Margaret O. Hyde

Chapters on various means of destroying the human brain -- from mad cow to Alzheimer's, Parkinson's to stroke, accidental injury to drugs and nutrient deficiency. Includes personal stories. -www.amazon.com/exec/obidos/ASIN/0531115437/top100

Neocortical Epilepsies
by Peter D. Williamson

Conference papers from 1998 on anatomy, physiology, and psychology in the normal and epileptic brain. -www.amazon.com/exec/obidos/ASIN/0781718724/top100

Developmental Variations in Learning: Social, Executive Function, Language, and Reading Skills
by Dennis L. Molfese