Spectral peak analysis and intrinsic neural timescales as markers for the state of consciousness

Abstract

Resting state EEG in patients with disorders of consciousness (DOC) is characterized by an increase of power in the delta frequency band and a concurrent decrease in the alpha range, equivalent to a weakening or disappearance of the alpha peak. Prolongation of Intrinsic Neural Timescales (INTs) is also associated with DOCs. Together, this raises the question whether the decreased alpha peak relates to the prolonged INTs and, importantly, how that can be used for diagnosing the state of consciousness in DOC individuals. Analyzing resting state EEG recordings from both healthy subjects and DOC patients, we measure INTs through autocorrelation window (ACW) and utilize peak analysis to quantify the weakening of the alpha peak. First, we replicate previous findings of prolonged ACW in DOC patients. We then identify significantly lower alpha peak measures in DOC compared to controls. Interestingly, spectral peaks shift from the alpha to the theta range in several DOC subjects while such change is absent in healthy controls. Next, our study reveals a close relationship between ACW and alpha peak in both healthy and DOC subjects, a correlation that holds for theta peaks in DOC. Further, the prolonged ACW correlates with the state of consciousness, as quantified by the Coma Recovery Scale-Revised (CRS-R), and mediates the relationship between theta peak and CRS-R. Finally, through split analyses and machine learning, we show that ACW and alpha peak measures conjointly distinguish healthy controls and DOC patients with high accuracy (95.5%). In conclusion, we demonstrate that the prolongation of ACW, together with spectral peak measures, holds promise to serve as additional EEG biomarkers for diagnosing the state of consciousness in DOC subjects.

Keywords: Alpha peak; Diagnosis; Disorders of consciousness; EEG; Intrinsic neural timescales.

Fig. 1

From EEG to Auto-Correlation Function and Power Spectral Density. Starting from an EEG signal (A) we measure Intrinsic Neural Timescales (INTs) using the Autocorrelation Function (B) and obtain the Power Spectral Density (PSD) via a Fast Fourier Transform (C). Panel (C) illustrates the measures we acquired through peak analysis.

Fig. 2

Variations in Power Spectral Density. Four distinct PSDs are displayed: a clear peak in the alpha range is visible for the healthy control (utmost left plot), which sees a considerable decrease in power, prominence and frequency in DOC patients (middle left plot), shifting to the theta range (middle right) or disappearing entirely (utmost right plot).

Fig. 3

Alpha spectral measures in controls and DOC patients. Through peak analysis we identify seven spectral measures, five directly related to the alpha peak and two non-peak related. Controls exhibit higher values across almost all measures, except for width (showing no statistically significant differences) and maximum power (higher in DOC patients). Ns: p ≥ 0.05; *0.01 < p < 0.05; ^**0.001 < p < 0.01; ***0.0001 < p < 0.001; ****p ≤ 0.0001.

Fig. 4

ACW, controls and DOC. Patients with Disorders of Consciousness exhibit prolonged ACW: this difference is clearly visible when comparing two Autocorrelation Functions side by side (A). In our study we replicate this finding and extend it to ACW-e⁻¹ (B). *0.01 < p < 0.05; ^**0.001 < p < 0.01; ***0.0001 < p < 0.001; ****p ≤ 0.0001.

Fig. 5

Correlations between ACW and alpha spectral measures in controls and DOC. ACW and alpha peak hold a strong relationship, especially for power, power patio and maximum power, which exhibit very strong correlations. These correlations are present both in healthy controls (A) and in DOC individuals (B). *0.01 < p < 0.05; **0.001 < p < 0.01; ***p < 0.001.

Fig. 6

Mediation analysis. Our mediation analysis starts from observing strong correlations between ACW and peak measures (Fig. 5) and correlations between ACW and CRS-R, which in turn, are absent between peak and CRS-R (A). The mediation models (B) hint at ACW possibly serving as mediators between peak and CRS-R. Panel (C) displays the strong correlation between power ratio and ACW, with CRS-R showing a weak tendency to decrease as ACW increases.

Fig. 7

ACW splits. The link between ACW and spectral peaks in DOC subjects extends from the weakening to the disappearance of all spectral peaks: patients in the top quantile for ACW length possess, on average, a lower chance of displaying a spectral peak (A). ACW allows segregation of peak measures and CRS-R (B). Ns: p ≥ 0.05; *0.01 < p < 0.05; **0.001 < p < 0.01; ***0.0001 < p < 0.001; ****p ≤ 0.0001.