Electrophysiological Frequency Band Ratio Measures Conflate Periodic and Aperiodic Neural Activity

Abstract

Band ratio measures, computed as the ratio of power between two frequency bands, are a common analysis measure in neuroelectrophysiological recordings. Band ratio measures are typically interpreted as reflecting quantitative measures of periodic, or oscillatory, activity. This assumes that the measure reflects relative powers of distinct periodic components that are well captured by predefined frequency ranges. However, electrophysiological signals contain periodic components and a 1/f-like aperiodic component, the latter of which contributes power across all frequencies. Here, we investigate whether band ratio measures truly reflect oscillatory power differences, and/or to what extent ratios may instead reflect other periodic changes, such as in center frequency or bandwidth, and/or aperiodic activity. In simulation, we investigate how band ratio measures relate to changes in multiple spectral features, and show how multiple periodic and aperiodic features influence band ratio measures. We validate these findings in human electroencephalography (EEG) data, comparing band ratio measures to parameterizations of power spectral features and find that multiple disparate features influence ratio measures. For example, the commonly applied θ/β ratio is most reflective of differences in aperiodic activity, and not oscillatory θ or β power. Collectively, we show that periodic and aperiodic features can create the same observed changes in band ratio measures, and that this is inconsistent with their typical interpretations as measures of periodic power. We conclude that band ratio measures are a non-specific measure, conflating multiple possible underlying spectral changes, and recommend explicit parameterization of neural power spectra as a more specific approach.

Keywords: aperiodic neural activity; frequency band ratios; neural oscillations; spectral analyses; θ/β ratio.

Figure 1.

Literature analysis of band ratio related articles. A, Associations between published journal articles referring to band ratio measures and cognitive and clinical associations. Each cell represents the proportion of articles referring to a specified band ratio measure that also mentions the corresponding association term. B, Total counts of the number of articles mentioning each band ratio measure.

Figure 2.

Overview of band ratio measures and spectral parameters. A, An example power spectrum in which shaded regions reflect the θ band (4–8 Hz) and β band (20–30 Hz), respectively. Band ratio measures, such as the θ/β ratio, are calculated by dividing the average power between these two bands. B, An example of a parameterized power spectrum, in which aperiodic activity is separated from measured periodic components. This is an example spectrum from EEG data, in which peaks in θ, α, and β power are present. C, Examples of simulated power spectra with and without component oscillations of the θ/β ratio. Black lines indicate the simulated data, with red line reflecting the model fit, the dashed blue line indicating the aperiodic component of the model fit, and the green lines indicating the location of canonical θ and β oscillations. Band ratio measures, although intended to measure periodic activity, will reflect power at the predetermined frequencies regardless of whether there is evidence of periodic activity at those frequencies.

Figure 3.

Equivalent band ratio differences from distinct changes. Simulations demonstrating the underdetermined nature of band ratio measures. In each case, the power spectrum plotted in orange has the same difference of measured θ/β ratio, indicated as Δ TBR, from the reference spectrum, in blue. This difference in ratio can arise from changes in multiple different features of the data, including a shift in: (A) periodic parameters such as the center frequency, power, or bandwidth of oscillations, and/or from a shift in; (B) aperiodic properties of the data, in this case the aperiodic exponent. Differences in aperiodic activity can induce differences in measured band ratios, even without any periodic components present (bottom panel).

Figure 4.

Single parameter simulations. Simulations of changes in measured θ/β ratio (TBR) as individual parameters are varied, including: (A) periodic parameters and (B) aperiodic parameters. Changes in θ center frequency show an increase in θ/β ratio as the heightened activity is better captured in the canonical band, then decreases as activity leaves the band. Increasing θ power and bandwidth both increase θ/β ratio while increasing β power and bandwidth decreases θ/β ratio. The center frequency and bandwidth of α peaks also influences measured θ/β ratio, although α is not supposed to be included in the measure. β parameters essentially have the inverse effect of changes in θ parameters. Changes in aperiodic exponent also substantially impact measured θ/β ratio, although offset has no effect. Note that the layout of this figure corresponds to Figure 3, in which examples of how each parameter influences measured θ/β ratio can be seen. CF: center frequency, PW: power, BW: bandwidth.

Figure 5.

Interacting parameter simulations. Measured θ/β ratio values in simulations as two spectral parameters are varied together. Ratio measures are plotted in log10 space because of their skewed distributions. Combinations plotted are aperiodic exponent with low band center frequency (A), as well as with low band power (B), and high band power (C). All combinations of varying parameters influence measured band ratio values.

Figure 6.

Correlations between spectral parameters and band ratio measures in EEG data. In a large EEG dataset, correlation results are reported for band ratios as compared with the periodic (left) and aperiodic (right) parameters for the (A) θ/β ratio, (B) θ/α ratio, and (C) α/β ratio. In A, these results show that the θ/β ratio is most strongly correlated with the aperiodic exponent, and less related to power in the θ or β. In contrast, B, C, show that any ratio measure that includes an α band is most strongly correlated to α power, meaning any α ratio is mostly reflecting just α power. CF: center frequency, PW: power, BW: bandwidth.

Figure 7.

Topographies of band ratio measures and spectral parameters. Topographical maps of the (A) ratios measures, including the θ/β ratio, θ/α ratio and α/β ratio. For comparison, the topographies of the aperiodic exponent (B) and of α power (D) are also presented. Each topography is scaled to relative range of the data, with higher values plotted in lighter colors (yellow). C, The spatial correlation between topographies of each ratio measure to spectral parameters including power of θ, α and β, and the aperiodic exponent (EXP). TBR: theta / beta ratio, TAR: theta / alpha ratio, ABR: alpha / beta ratio.

Figure 8.

Changes in ratios and spectral parameters across blocks. Each row reflects a band ratio measure and each column reflects a spectral parameter. Each point is a difference measure across blocks, the value of the measure in a block, minus the value of that measure in the prior block, collected across all subjects. Printed in the inset is the spearman correlation between the measures. Consistent with prior analyses, changes across blocks in the θ/β ratio are most correlated with changes in aperiodic exponent, and changes in θ/α and α/β are most correlated with changes in α power.