Here’s a new paper by Mike X Cohen that I would like to share given that he brings out a topic that has recently intriguied me and gives a potential solution for it.
BioRxiv (2020) – DOI: doi.org/10.1101/2020.07.09.195784.
A data-driven method to identify frequency boundaries in multichannel electrophysiology data
Micheal X Cohen
The following tweet it’s a good way to start talking about this paper:
Cohen basically states the following points in the paper:
- The brain operates at different rhythms and numerical boundaries of those frequencies are kind of subjective in the literature. For example, we find papers in which the alpha-band is defined within 7-12 Hz, 8-13 Hz or 9-11Hz, and in others, the alpha-band is divided in sub-bands (low: 8-12 Hz; high: 10-12 Hz). A metaphor would be the figure from this post. If you take a look at the figure you’ll see that it’s hard to tell when where would you set the colour-boundaries to separate the different spectrum of colours. A similar thing happens with the power spectrum of EEG data and frequency boundaries.
- In order to provide a solution for this, Cohen proposes an empirical (data-driven multivariate) method to identify spectral boundaries based on spatiotemporal patterns across a range of frequencies. This method is termed getBounds and it works like this:
gedBounds identifies patterns in covariance matrices that maximally separate narrowband from broadband activity, and then identifies clusters in the correlation matrix of those spatial patterns over all frequencies, using the dbscan algorithm. Those clusters are empirically derived frequency bands, from which boundaries can be extracted.
- The method is fast, deterministic, and easy to implement (MATLAB code is provided), and is appropriate for multichannel EEG data in which narrowband activity can be expected.
For more details about the method I invite you to read the whole paper. As always, Cohen explains such in detail everything. And, he alos provides the MATLAB code to generate the simulation and implement gedBounds can be found here. and you can even try it with your own data!
Welcome to neuroscience, gedBounds!
Feature image from Pexels – C0 license.