First, two things:
Julio (@ArcadeShrimp) did a ton of the work on this one. He is excellent and you should follow him, and hire him for your lab.
This project is a fully open - code, data, all annotated & documented, is available:
https://github.com/voytekresearch/BandRatios …
/2
-
-
Prikaži ovu nit
-
A while ago, I became interested in BandRatio measures - in which one takes the ratio of power across two frequency bands. These include measures like the theta / beta ratio used in a lot of cognitive and clinical work (for example, with ADHD). /3pic.twitter.com/FDac79heXR
Prikaži ovu nit -
These measures are common - we ran a literature analysis (Python tool: LISC), finding over 250 papers mentioning them in abstracts alone. They are used across cognitive, clinical and neurofeedback studies. Plotted is proportion of papers across topics, column is # of papers. /4pic.twitter.com/Q0AlT956ui
Prikaži ovu nit -
Band ratios are typically used and interpreted as reflecting oscillatory power. But do they? For example, what if there is no oscillatory power? Sometimes there is no power over and above the aperiodic, which we consider meaning no oscillation. What happens then? /5pic.twitter.com/mDiLr7Ny7b
Prikaži ovu nit -
Mostly I work on 'parameterizing' power spectra, to separate and measure periodic activity (oscillations) and aperiodic activity. This means we can quantify aperiodic (blue line) and oscillations (overlying peaks, in red). We use this to simulate & quantify power spectra. /6pic.twitter.com/2byMsadeyu
Prikaži ovu nit -
First, we ran a whole bunch of simulations. We are able to show how lots of different changes in the data can affect band ratio measures. For example, every comparison here has the same difference in theta / beta ratio - from many different spectral changes. /7pic.twitter.com/iVdaUAKhgO
Prikaži ovu nit -
There are tons of more simulations in the paper, that dive into this more, and quantify how much each feature affects band ratios. Next: what happens in real data? For this we used an openly available dataset (many thanks to the
@ChildMindInst institute for releasing it!) /8Prikaži ovu nit -
In the EEG data, we calculated band ratios and parameterized spectral features, and check which spectral features band ratios relate to. Result: the theta / beta ratio (which *should* track theta and beta power) is actually much more related to aperiodic activity. /9pic.twitter.com/ipBw176o6a
Prikaži ovu nit -
Another way to look at it - what are the spatial profiles? We see, for example, that the topography for theta / beta ratio is very similar to that of aperiodic exponent. Also, ratios with alpha also seems to basically tracks alpha. /10pic.twitter.com/eJhmLRRTfz
Prikaži ovu nit -
Findings recap: Band ratio measures, as proposed and interpreted, *should* reflect oscillatory power. We show many different spectral features can affect their measurement. In real data, they often related to aperiodic activity, or to power in only one band. /11
Prikaži ovu nit -
What does this mean? We conclude band ratio measures are non-specific, and do not specifically reflect oscillatory activity. Band ratio measures should not be uncritically interpreted as reflecting patterns of oscillatory power. /12
Prikaži ovu nit -
What next? I think we should no longer use band ratio measures, and that we don't need them. Want to know what is changing in the data, in what ways, and what it correlates to? Parameterize the spectra. Compare those measures, knowing exactly what changed. /13
Prikaži ovu nit -
I think this means the prior literature on band ratios needs to be interpreted carefully. In particular, the theta / beta ratio often likely reflects aperiodic activity. What's affecting ratios could vary across studies though, so care should be taken to check case-by-case. /14
Prikaži ovu nit -
If you have (or are) using band ratios, this is probably not great news. But I hope spectral parameterizations offer a productive way forward. My optimistic view is that hopefully this offers a new approach that might clarify some of the inconsistencies in the literature. /15
Prikaži ovu nit -
If you want to try our approach to parameterizing neural power spectra, it is available as an open-source tool. If you need any guidance using it, especially if you want to try it with or instead of band ratio measures - please get in touch. /16https://github.com/fooof-tools/fooof …
Prikaži ovu nit -
There is more stuff in the paper! Also, a reminder that this is a fully open project. You can check out exactly what we did, and how we did it. We spent a lot of time to make this completely shareable (thanks to Julio), so please take advantage! /17https://github.com/voytekresearch/BandRatios …
Prikaži ovu nit -
Bonus: simultaneously, another group started investigating if the theta / beta ratio captured aperiodic activity in the context of ADHD. We helped a bit with their paper showing that it does! A nice convergence & demonstration in clinical data. /end https://www.physiology.org/doi/full/10.1152/jn.00388.2019 …
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

. he/him.