Do Neural Networks Segregate Information By Frequency?

Can information be directed to different networks in the brain depending on the "transmission frequency", like the channels on a TV? A 2006 Cerebral Cortex paper reveals that this may not be as absurd as it sounds.

A relatively new technique in cognitive neuroscience is the use of frequency tagging, where a stimulus (whether visual or auditory) is presented at a certain rapid frequency, perhaps onsetting and offsetting six times per second (6Hz). A second stimulus may be presented at 4.5Hz. The frequencies can then be detected in the brain using magnetic or electric methods (MEG or EEG), and one can track which parts of the brain seem to be processing the frequency-tagged stimulus when it is attended, recalled, ignored, etc.

Authors Ding, Sperlin & Srinivasan say that the method is a little more complex than that. They argue that by tagging a stimulus with a certain frequency - e.g., 6Hz - you may be biasing the brain to process that stimulus with populations of neurons which are capable of synchronizing to that frequency. So, in a way, frequency tagging is actually an invasive method, and all conclusions based on it about how the brain processes various stimuli are necessarily reflective of the frequency tags used.

To test this idea, the authors gave subjects a target detection task, in which they had to indicate when one of many white circles in a display suddenly became a triangle. These white circles were arranged inside each of two "rings", one red and one green, nested inside one another; only one ring was ever relevant on each trial, and subjects were told to ignore the irrelevant ring for that trial. Critically, the white circles lying on one ring (either the relevant or irrelevant one) would be tagged with one of 15 frequencies (from 2 to 20 Hz), while the white circles lying on the other ring would be tagged with a random and continually changing frequency (and should therefore average to noise). Subjects completed 60 trials while their scalp electrical activity was measured with a 128 electrode EEG net.

Subjects did quite well at this simple task, as you might expect. The interesting data comes from the EEG, which was fourier transformed to extract the power of the EEG response at each flicker frequency. The authors calculated the amount of increase in EEG power at the flicker frequency when the attended ring was flickering at that frequency, relative to when it was flickering at random, at each of the 128 electrodes for each of the 15 frequencies.

The results showed that EEG power at the flicker frequency was generally higher when the attended ring was the one that flickered (as you would expect). Contrary to expectations, however, when the attended ring flickered at 8 or 9.2 Hz, the EEG power spectrum showed that this frequency was suppressed relative to when the unattended ring flickered at those frequencies. Here, we begin to see that the flicker tagging method is far more complex than considered previously.

In addition, the location of these effects changed depending on the frequency. For lower frequencies (2.5 Hz) frontal electrodes showed significant effects; in contrast, power at parietal electrodes was maximized with flicker frequencies of 9.2. Unfortunately, the authors did not use source localization to confirm that these differences reflected different anatomical networks, but it seems clear that there is some mechanistic difference in the way these irrelevant frequencies are processed, even when the task itself remains the same.

The authors conclude that these results suggest "the presence of cortical resonance phenomena that depend on both spatial location and input frequency. Thus, by changing the flicker frequencies in a frequency-tagging experiment we potentially select for different functional networks whose natural frequencies match the flicker frequency." They also further speculate that an occipital-frontal network may be preferentially engaged by attended delta (2-4 Hz) and upper alpha (10-11 Hz) frequency tags, whereas parietal and posterior frontal regions may respond more to unattended flicker in the lower alpha band (8-10 Hz).

To my knowledge there is relatively little reason to suspect that any region of the brain, much less parietal and frontal cortex, would appear more strongly affected by frequency tagging when those items are unattended. Cortical resonance as a whole is still a highly mysterious topic, and only highly simplistic theories exist as to how attention may interact with those resonant frequencies. Perhaps most damningly, this study demonstrates that previous work with frequency tagging can no longer be cleanly interpreted, as it's unclear to what extent those results are based on the precise frequency bands that were chosen.

On the bright side, this study does provide fascinating ideas for future research. Can other methods also reveal characteristically resonant frequency bands in various networks? Might this be a good way of dissociating networked but distinct mechanisms? And finally, how does attention affect these frequencies?

More like this

In this poster, Bastos, Mullen and colleagues show that they can analyze electrical oscillations on the scalp of human subjects and predict how quickly they will respond in a simple target detection task. They do this by an interesting method known as the Steady State Visual Evoked Potential (…
Intuitively, an adaptive information processing system should deal with unique or unusual information in a special way. For example, an unusual encounter might indicate that an organism's environment is changing, and by implication that there's a new potential for danger. Or novel information can…
More than 100 years ago, Sigmund Freud, the father of psychoanalysis, proposed a mechanism called repression, whereby desires and impulses are actively pushed into the unconscious mind. For Freud, repression was a defence mechanism - the repressed memories are often traumatic in nature, but,…
Hallucinations are often associated with psychiatric conditions such as schizophrenia or with LSD and related drugs. Hearing voices is a characteristic symptom which is reported by about 70% of schizophrenic patients, as well as by some 15% of patients with mood disorders such as depression; and…