Lab Talk

BandRanges

The Remarkable Inconsistency of EEG Frequency Band Definitions

Defining the power spectrum in terms of different ranges named delta, theta, alpha, beta and gamma forms a fundamental framework in the EEG literature today.  Yet these band definitions have a shockingly wide variability in the literature.

Since the 1930s, when frequency oscillations were first measured by Hans Berger, researchers have typically analyzed the EEG power spectrum by slicing it into a small number of frequency “bands”. These predefined spectral windows have been given the labels of delta, theta, alpha, beta, and gamma, moving from the lower to the upper end of the spectrum respectively. The bands were initially defined based on constraints of technology and analysis of the 1930s. They were then perpetuated by the mechanical band pass filters developed by individuals such as William Walter Grey and Edmund Kaiser (see Down a Rabbit Hole? A History of EEG Analysis) to simplify the analysis of the power spectrum, and continue to prevail today in modern EEG research.

Given these bands form a fundamental backbone of the EEG literature today, one important consideration, especially when comparing across EEG studies, is whether the frequency definition of these bands is being systematically applied across studies. In other words, does one researcher’s “alpha” or “beta” mean the same, from the perspective of the power spectrum, as another researchers? And if researchers are not using the same frequency banding, what magnitude of variability and confusion are we talking about?

To explore this, we reviewed 135 resting-state EEG studies that looked at frequency bands in relation to mental health disorders. The papers spanned 10 disorder types including ADHD, schizophrenia, depression, OCD, PTSD, anxiety, panic disorder, bipolar disorder, autism and addiction, to pull out the reported frequency ranges for each of the bands described in each study.

To start with, it is important to note that not all 135 studies included all 5 major bands in their analysis. For example only 20% of the studies we reviewed actually analyzed the gamma band in contrast to 84% analyzing alpha and theta bands. In addition, some studies split the major bands into sub-bands, giving them labels such as beta1 or alpha2. To allow us to standardize our comparison across studies we collapsed across these sub-bands, and considered the start of each major band to be the lower end of the lowest frequency sub-band (e.g. the starting frequency of beta1), and the end of the band to be the upper end of the highest frequency sub-band  (e.g. the ending frequency of beta3).

 The result of our analysis shows that there is shocking variability across studies in terms of the frequency windows used to define each band, especially towards the upper end of the spectrum.  The image at the top of this post shows the most common range for each band as a wide bar and the entire possibility of ranges as a thinner line.

Delta shows wide variability in its range with some researchers choosing to use a narrow window (e.g. 0.5-2.5Hz) whilst others choose to use a broader window (e.g. 1.5-6Hz) with the most common range being between 1.5-3.5Hz. Some of this may be on account of differences in inbuilt filters used across different EEG hardware that constrain the lower end. The level of consistency is slightly better for alpha and theta bands.  For theta, the most common windows are 3.5-7.5Hz or 4-8Hz, used in 25% and 29% of studies respectively. However, across the reviewed studies the theta band started anywhere between 2.5Hz to 6.5Hz indicating that there is considerable overlap with the delta band across studies. The frequency of papers using each start and stop point for each band (excluding gamma which is sparsely represented ) is shown below.

 

Alpha generally begins between 7.5 and 8.5 Hz with 8 Hz being most common, but its end point had much larger variability. The most common windows are 8-12Hz or 8-13Hz  (used in 21% and 35% of studies respectively) and, with one exception, the remaining studies don’t stray beyond 7.5 or 8.5Hz at the lower end or beyond 11 or 14Hz at the upper end of the frequency window.  The variability in the beta band however, is immense.  The most common frequency window for the beta band is between 12 or 13Hz  and 25 or 30Hz. However some researchers consider beta to start instead at 16 or 18Hz or end instead at 50Hz. This means that the upper frequency values for beta vary by as much as 30Hz (from 18 to 50Hz).  Similarly, some studies define the gamma band as being between 30-40Hz, whilst others define it as being between 30 and 100 Hz (not included in the figure). This means that the upper frequency value used to compute gamma activity across the reviewed studies can vary by as much as ~60Hz (from 38-100Hz). The combined variability at the lower end of the gamma band (from 20-37Hz) and the upper end of the beta band also means there is considerable overlap between the beta and gamma bands from one study to the next.

see related post The Blue Frog in the EEG

Overall, these results provide an insight into the confusion and inconsistency which underlies the definition of the frequency bands used as result markers in so many EEG studies. Overlap between the bands, and the considerable variability in the start and end points of the banding across studies calls into question the validity of making systematic comparisons across studies. And it categorically confirms that you can’t always take for granted that what one researcher means when they report results for a particular band is going to be the same as another’s. Beyond the lab, considering the range of EEG equipment and methods used even in a clinical context, this offers strong caution in any clinical interpretation based on reports in the literature. For instance, if the beta band can have a full 30 Hz range of variability, what exactly is the theta/beta ratio that even has FDA approval as a diagnostic marker?

see related post ADHD and the Theta/Beta ratio

5 thoughts on “The Remarkable Inconsistency of EEG Frequency Band Definitions

      1. Any updates on the publication? I’m looking for a peer reviewed paper on the subject and your work here is the best I’ve seen. I’d love to be able to cite it.

Leave a Reply