High Variability Periods in the EEG Distinguish Cognitive Brain States

Abstract

Objective: To describe a novel measure of EEG signal variability that distinguishes cognitive brain states.

Method: We describe a novel characterization of amplitude variability in the EEG signal termed "High Variability Periods" or "HVPs", defined as segments when the standard deviation of a moving window is continuously higher than the quartile cutoff. We characterize the parameter space of the metric in terms of window size, overlap, and threshold to suggest ideal parameter choice and compare its performance as a discriminator of brain state to alternate single channel measures of variability such as entropy, complexity, harmonic regression fit, and spectral measures.

Results: We show that the average HVP duration provides a substantially distinct view of the signal relative to alternate metrics of variability and, when used in combination with these metrics, significantly enhances the ability to predict whether an individual has their eyes open or closed and is performing a working memory and Raven's pattern completion task. In addition, HVPs disappear under anesthesia and do not reappear in early periods of recovery.

Conclusions: HVP metrics enhance the discrimination of various brain states and are fast to estimate.

Significance: HVP metrics can provide an additional view of signal variability that has potential clinical application in the rapid discrimination of brain states.

Keywords: anesthesia; brain states; electroencephalography (EEG); entropy; working memory.

Figure 1

Definition of High Variability Periods (HVP). (A) Mean standard deviation of amplitude (A_SD in µV) across individuals for windows of different sizes. Error bars indicate standard error across individuals. (B) Resting state EEG signal from one person with amplitude shown in multiples of the standard deviation (A_SD, grey) along with the SD of the signal amplitude estimated for varying window sizes from 5, 10, and 15 s (colored lines). Dashed line represents the SD of the signal amplitude of the whole trace, i.e., A_SD or 1 SD of the amplitude. (C) A_SD using 15 s window and 50% overlap for the EEG signal, as shown in (A). The dashed line represents the 25th percentile of the moving A_SD values, which serves as the threshold T. All segments >T are defined as High Variability Periods or HVPs. Segments consistently <T are defined as low variability periods or LVPs. Durations are computed as the time between the threshold crossings. Area (for HVPs only) is computed as the sum of all SD values across the duration. (D) Mean standard deviation of amplitude (A_SD in µV) across 2 monkeys and 4 recordings for windows of different sizes. Error bars indicate standard deviation across monkeys. (E) Similar resting state signal along with SD estimates for different moving window sizes as shown in (A), but from a monkey rather than human. (F) The moving A_SD with a 15 s window and 50% overlap for the EEG signal, as shown in (C).

Figure 2

Parameter choices and HVP metrics for resting state eyes closed (EC). (A) Mean HVP and LVP durations across individuals for different window sizes (x-axis) and window overlaps (see legend) when threshold is maintained at the 25th percentile of the A_SD trace. (B) Mean HVP and LVP durations across individuals for different window sizes (x-axis) and thresholds (see legend), when window overlaps, is maintained at 50%. (C) Mean ratio of HVP and LVP durations across individuals for different window sizes (x-axis) and window overlaps (see legend) when threshold is maintained at the 25th percentile of the A_SD trace. (D) Mean ratio of HVP and LVP durations across individuals for different window sizes (x-axis) and thresholds (see legend) when window overlaps are maintained at 50%.

Figure 3

HVP metrics across behavioral conditions. (A) HVP duration (HVP-D; window overlap of 50% and 25th threshold percentile) across all 28 individuals (mean ± SEM) as a function of window size for resting eyes closed (EC; closed circle) and eyes open (EO; open circle) and for working memory (EO-WM) and pattern completion (EO-PC) tasks. (B) HVP area (HVP-A; mean ± SEM) as a function of window size for the same tasks/conditions as in (A). (C) Mean ratio of HVP and LVP duration as a function of window size for the same tasks/conditions as in A. (Error bars not shown for readability. SEM values are ~±0.2 to ±0.6). (D) HVP rate (mean ± SEM) as a function of window size for the same tasks/conditions as in (A).

Figure 4

HVPs in monkeys and impact of anesthesia. (A) Mean HVP duration across 4 recording sessions across 2 blindfolded monkeys as a function of window size (overlap of 50% and threshold at 25th percentile). (B) Mean HVP area as a function of window size (overlap of 50% and threshold at 25th percentile). (C) Mean HVP-D/LVP-D ratio as a function of window size (overlap of 50% and threshold at 25th percentile). (D,E) Examples from a single channel in each of two monkeys showing the EEG signal (gray, normalized by the SD) before (rest) and after injection of ketamine (low-anesthetic and deep-anesthetic) and following injection of an antagonist for reversal of the anesthetic effect (recovery). The moving A_SD used to calculate HVPs is shown in red. The recording in (A) used a resting state threshold of 280 µV, and the recording in (B) used a resting state threshold of 153 µV. (F–H) HVP duration, area, and rate per minute in each state of anesthesia. HVP events completely disappeared in deep anesthesia and did not reappear in the initial recovery phase.