Do alpha-power and SPEX topographical changes correlate between eyes-open and eyes-closed?

[MakotoMiyakoshi]

Posting my old unpublished observations.

I used Dortmund datasets by Edmund Wascher. n=608, 64 channels, eyes open and closed.
https://www.nature.com/articles/s41597-024-03797-w#libraryItemId=17697190

Hmm... they do not overlap. Interesting, isn't it?

Is there anyone who wants to write a paper on this?

[Contrerana]

I started to look into something similar (task vs resting state) with infants EEG data, but as maturation it is a big confound here it is difficult to make any conclusions. I will be happy to work with you in adults first, less variability than in infants EEG, and them move back to infants...

[MakotoMiyakoshi]

Hi @Contrerana,

There are multiple factors that affect the 1/f-ness, but as far as we know, myelination is one of the major ones. An unmyelinated axon is effectively treated the same as a dendrite and is therefore subject to the dendritic filter” so I see your concern. However, I do not think it means it is meaningless to study it--you can still study, for example, within-subject comparison, which Mate has been saying the most reliable method we can use.

Anyway, if you want to work on adult data, you are welcome.

[Contrerana]

Agreed, myelination and changes in the E/I balance are key factors during development that will affect 1/f. But also, how we preprocess the data (infant artefacts are very different to adults) and the bands of interest (e.g infant alpha is more in the range of 6 to 9Hz) also likely to affect how we calculate/ interpret the results. That is why I think it is easier first achieve a consensus in the meaning and "good practices" level in adults and then move towards infants. Infant data is messier and much more variability, even more than in adults. But I may be totally wrong too, lots of experience in EEG but newly arrived in the 1\f topic.

[dieloz]

Happy to chime in!
@MakotoMiyakoshi, if you could provide a bit more information about the power scale so we can properly interpret the negative values (and also about the SPEX) that would be great.

It seems that the Dortmund dataset may contain differences in sensorimotor mu that can be detected more or less easily depending on the magnitude of overlapping classical alpha. This relates to the additive vs. multiplicative signal-generation discussion raised in the email list. Alpha estimates during eyes-closed conditions may bias (or not) the 1/f estimation relative to eyes-open. Simply closing the eyes can shift the peak frequency in some cases (https://www.eneuro.org/content/7/1/ENEURO.0268-19.2019), so may be these are two different beasts.

Not relevant for the Dortmund dataset, but closing the eyes and performing a tactile task places participants in a very different brain state compared to performing the same task with eyes open (https://www.eneuro.org/content/9/1/ENEURO.0412-21.2021).

I’m very happy to contribute to study how these brain-state changes affect both oscillatory components and the 1/f exponent. There are open datasets with re-test measurements (cross-over designs) that we could use to examine how reliable these differences are (i.e., the trait vs. state debate).

[MakotoMiyakoshi]

The PSD color scales are in 10*log10 microV^2/Hz (dB). 0dB represents 1 microV^2/Hz in linear scale.

In fact, it is strange not to see 'red' in the eyes open condition. I'll ask Hyeonseok about it. Please wait for update.

[EugenMasherov]

1. Is the power in the figures on the left on a logarithmic scale?
2. What algorithm was used to remove the influence of the alpha rhythm peak with eyes closed?

And a simple slope estimation algorithm. Given a spectrum p(w), we calculate the slope for all possible frequency pairs s=(ln(p(w1)-ln(p(w2))/(ln(w1)-ln(w2)) and take the median of the resulting values. Peak values ​​will yield erroneous estimates, but the median, as a robust estimator, will discard them. It would be interesting to compare this approach with others.

[MakotoMiyakoshi]

1. Yes, 10*log10 microV^2/Hz (dB).
2. FOOOF with a single-peak detection (default is 3 peaks) See Donoghue et al. (2020)

[MarjanZamani1990]

From my view, the dissociation between the spatially focal modulation of alpha power and the globally distributed changes in the aperiodic spectral exponent across eyes-open and eyes-closed conditions appears to support the idea that rhythmic activity is more closely related to local circuit dynamics, whereas the aperiodic component may index global, state-dependent properties of large-scale neural networks.

Periodic vs. Aperiodic Components in EEG
From Local Synchronization to Global Network State

In my opinion, one of the fundamental questions in EEG analysis concerns the physical and physiological origins of periodic components (rhythms) and the aperiodic background (1/f-like spectral activity), and why these components seem to behave differently across conditions such as age, resting state, task engagement, or pathology.

A simple yet potentially useful conceptual framework is to view the brain not as a single generator, but as a collection of local regions, each comprising neuronal populations with their own intrinsic dynamics. Under specific conditions, activity within these populations may become partially synchronized, giving rise to dominant, band-limited oscillations. These oscillations are not ideal sinusoids; rather, they are noisy, transient, and circuit-dependent processes that appear as periodic peaks in the power spectrum—for example, posterior alpha during eyes-closed rest.

In contrast, scalp EEG appears to reflect the superposition of activity from many such regions. These regions differ in their dominant frequencies, are not necessarily phase-aligned, exhibit variable amplitudes and coupling strengths, and include both oscillatory and non-oscillatory activity. As a result, the time-domain signal may appear unstructured, while in the frequency domain it often exhibits a 1/f-like spectral background with superimposed local peaks.

From this perspective, the aperiodic component should perhaps not be interpreted as meaningless noise, but rather as a statistical signature of aggregated neural activity across multiple spatial and temporal scales. This framework may help explain why rhythmic components tend to be local and circuit-specific, whereas the 1/f spectral slope appears to be a more global property, potentially sensitive to overall network state, arousal level, excitation–inhibition (E/I) balance, age, and task demands.

Importantly, the aperiodic component may not be merely an emergent consequence of signal summation. It may also have well-defined biophysical origins. At the cellular level, cable theory and dendritic geometry impose low-pass filtering properties that can naturally give rise to 1/f-like spectra. At the synaptic and network levels, E/I balance, changes in firing rates, and proximity to critical or near-critical dynamics may directly modulate the spectral slope. Additionally, heterogeneity in time constants, multiple spatial and temporal scales, and network architecture may promote statistical self-similarity and power-law behavior.

This perspective also allows for a more nuanced interpretation of the idea that rhythms are associated with “rest” or low-input states. While certain rhythms (such as alpha) are more prominent during rest or reduced sensory input, this does not necessarily imply that they reflect an idle brain. Local synchronization can occur in both rest and task states; however, during rest, reduced competing inputs and network perturbations may allow these oscillations to become more stable and spectrally pronounced. In contrast, during active task engagement, complex network dynamics may be more strongly reflected in changes to the aperiodic background and the 1/f slope.

From this viewpoint, spectral analysis of EEG can be seen as a form of multiscale decomposition: a tool for revealing how the composite brain signal may be constructed from locally organized periodic processes and global, state-dependent aperiodic dynamics. Rather than offering a final or definitive model, this framework aims to provide a coherent conceptual language for interpreting the relationship between rhythmic activity and spectral slope in EEG.

[MakotoMiyakoshi]

Hi @Marjanz69,

I agree that it is surprising how the difference in spike width kind of 'penetrates' layers of spatial scales so that we can still distinguish it in scalp EEG in the frequency domain. How the difference survive and go through the layers of spatial scale difference is not entirely clear, which was why the developers of FOOOF validated it very carefully.