Short communication
Assessing transient cross-frequency coupling in EEG data

https://doi.org/10.1016/j.jneumeth.2007.10.012Get rights and content

Abstract

Synchronization of oscillatory EEG signals across different frequency bands is receiving waxing interest in cognitive neuroscience and neurophysiology, and cross-frequency coupling is being increasingly linked to cognitive and perceptual processes. Several methods exist to examine cross-frequency coupling, although each has its limitations, typically by being flexible only over time or over frequency. Here, a method for assessing transient cross-frequency coupling is presented, which allows one to test for the presence of multiple, dynamic, and flexible cross-frequency coupling structure over both time and frequency. The method is applied to intracranial EEG data, and strong coupling between gamma (∼40–80 Hz) and upper theta (∼7–9 Hz) was observed. This method might have useful applications in uncovering the electrophysiological correlates of cognitive processes.

Section snippets

Dataset

The data analyzed here were taken from a patient (female, aged 37 years) with electrodes implanted for pre-surgical evaluation of epilepsy. Electrode placement was made on clinical grounds. The electrode used here was in a 4 × 8 grid placed over left lateral temporal cortex; the electrode used is located in the posterior superior temporal gyrus. During the recording, the patient watched a 4-s video clip of two solid circles moving on the screen. The task was to pay attention and to judge, on a

Brief overview of method

The idea behind this approach is that if the power of oscillations at a high frequency are synchronized with a slower frequency oscillation, the upper frequency power time series will itself oscillate at that lower frequency (Bruns et al., 2000). Here, an empirical approach is used to select the lower frequency band based on the observed fluctuations in the upper power time series, and standard phase coherence measures are used to evaluate the phase synchronization between the upper frequency

Details of method

All analyses were conducted in Matlab 6.5 using the signal processing and eeglab toolboxes, the latter of which is free to download (Delorme and Makeig, 2004), and supplemented by code written by the author. Sample Matlab code for conducting this analysis is available from the author upon request. This method utilizes a three-step plan, which is repeated over many time–frequency windows. Here, windows of 400 ms and 5 Hz were used; in the following example I use one of many windows to illustrate

Statistical tests

Statistics can be conducted in one of several ways, for example via data-based bootstrapping techniques in order to determine the distance between the observed SIm value and those expected by chance, or by transforming SIm values via the Fisher z-transform and entering into a standard statistical software package such as SPSS. The former method is useful for determining whether cross-frequency coupling is greater than would be observed by chance; the latter method is useful for parametric

Interpretation of results

From inspection of Fig. 2, it can be seen that significant cross-frequency coupling was observed in the gamma frequency range, especially around ∼40–80 Hz. The lower frequency of this coupling was generally around 7–10 Hz (i.e., upper theta to lower alpha). This concentration of cross-frequency coupling in this range is interesting in light of previous research: gamma oscillations have been observed in several regions of the brain, and have been linked to cognitive and neural processes, including

Limitations and conclusions

Practically, this procedure is time- and processor-intensive. Down-sampling in time and frequency, or using parallel clustered computer networks to run analyses are ways to decrease computation time. Physiologically, it is generally assumed that upper frequencies are synchronized to the lower frequency, although the inverse could occur as well. Which frequency band – if any – is inducing modulations in the other frequency band is not addressed here. Other methods exist to estimate causality in

Acknowledgements

I thank Juergen Fell and anonymous reviewers for comments on the manuscript, and Nikolai Axmacher and Thorsten Kranz for useful comments on the methods.

References (24)

  • R.T. Canolty et al.

    High gamma power is phase-locked to theta oscillations in human neocortex

    Science

    (2006)
  • J.J. Chrobak et al.

    Gamma oscillations in the entorhinal cortex of the freely behaving rat

    J Neurosci

    (1998)
  • Cited by (205)

    • Frequency band coupling with high-frequency activities in tonic-clonic seizures shifts from θ to δ band

      2022, Clinical Neurophysiology
      Citation Excerpt :

      For the statistical assessment of SIm, we shifted the phase-time series of the HFA amplitude and calculated the bootstrapped SIm (SImb) by using the δ or θ phase. This procedure was repeated 1000 times to create the distribution of SImb (Cohen, 2008), which represented the surrogate data. The maximum values of the distribution of SImb were stored at each surrogate data point, and the distribution of the maximum values was obtained.

    View all citing articles on Scopus
    View full text