Thread by @Tomdonoghue

New Preprint 🎉

"Methodological considerations for studying neural oscillations"

With Natalie Schaworonkow (@nschawor) and Bradley Voytek (@bradleyvoytek), we review key methodological issues and concerns for analyzing oscillatory neural activity.

🧵:
psyarxiv.com/hvd67/

We use simulated data to demonstrate and describe 7 key issues we think should always be considered for measuring neural oscillations.

We try and review and pull together recommendations, citing and combining topics from across the current literature.

#tweeprint #AcademicEEG

#1: Neural oscillations are not always present.

Neural activity contains aperiodic activity, which has power across all frequencies, and can appear rhythmic.

To validate oscillation specific power, analyses should start with a detection step, verifying oscillatory presence.

#2: Neural oscillations vary in peak frequency

Canonically defined band ranges may not match the observed data. This can lead to mis-estimations of measures of interest.

Analyses should verify center frequencies of examined oscillations, and individualize band ranges if needed.

#3: Neural oscillations co-exist with dynamic aperiodic activity.

Measured changes in the data may reflect dynamic aperiodic activity, rather than oscillations.

Analyses should explicitly measure & control for aperiodic activity, to evaluate if it potentially explains findings.

#4: Neural oscillations are temporally variable

Oscillations are often 'bursty', which can confound power measures. Averaged power may falsely look sustained across time.

Burst detection measures can assess temporal variability, and disentangle different changes in the data.

#5: Neural oscillations are non-sinusoidal

Variable waveform shape can confound sinusoid based measures, leading to mis-estimations of power and/or spurious coupling.

Waveform shape can be explicitly measured and controlled for, to examine if it's potentially impacting results.

#6 Multiple oscillations coexist across the brain

Variations in multiple underlying sources can combine & cancel such that such that electrode activity may not be representative of underlying sources.

Source separation approaches can be used to disentangle overlapping sources.

#7: Measures of neural oscillations require sufficient signal-to-noise ratio (SNR)

Analyses with low oscillatory SNR (low power) can be unreliable, giving false positives/negatives.

SNR should be evaluated to verify that there is sufficient signal power for reliable estimates.

That's it - those are our topics for when analyzing neural oscillations!

There are of course nuances, and lots of discussion in the paper, but to a first approximation we believe these topics should always be checked / considered.

The neural oscillation checklist:

Notably, this project is aimed at being a practical review (these topics are not at all novel observations) and I hope this project offers a useful guide & reference list to lots of great work and tools in this space!

Thanks to everyone doing important work in this area!

Also - we would love to hear any feedback!

If anyone has any comments or suggestions that they think we should add or mention please do let us know and we will do our best to integrate suggestions!

Also - code!

I'm a huge fan of methods-testing through simulation, which is something I highly recommend. In this project, we are mostly using NeuroDSP for the simulations (neurodsp-tools.github.io).

The code for this project is all openly available:
github.com/voytekresearch…

Okay, I think that's it!

Thanks again to my co-authors, to everyone who works on tools that enable these projects, and to everyone reading & sharing - the reception & feedback really does boost motivation to continue this work!

/end

Paper link, again:
psyarxiv.com/hvd67/