Influence of drugs on sleep EEG

Three overnight EEG recordings from a double-blind test of drug influence on nocturnal sleep were subjected to:

1. Spectral analysis in the bands related to sleep spindles and slow wave activity (SWA),
2. MP decomposition in stochastic dictionaries [3].

Columns in Figure 3 present, from left to right, results for the night after administrating placebo and two sleep inducing drugs. We analyzed artifact-free data from C3-A2 derivation of the 10-20 system, sampled at 128 Hz.

Figure 3: Analysis of sleep EEG recorded after administration of placebo (left column) and sleep-inducing drugs (middle and right columns). Rows a and c present spectra (red line) averaged in 4-sec. artifact-free epochs of stage II (sleep spindles in a) or stages III and IV (slow wave activity, SWA, in c). Numerical values on the right give the average spectral power and its mean square error. Rows b and d present structures from the MP parameterization of the same epochs, qualified as sleep spindles (b) or SWA (d). Each blue dot represents one structure of given frequency (horizontal) and amplitude (vertical), conforming to criteria set as: for sleep spindles frequency 12-15 Hz, time duration 0.5-2.5 sec., peak-to-peak amplitude above 15 $\mu V$, and for SWA frequency 0.5-4 Hz, duration above 0.5 sec., amplitude $> 75 \mu V$. Parameters listed below the power include: average number of detected structures per minute, total time of qualifying EEG, average (weighted by amplitudes) frequency fav, its standard deviation $v_f$ and average amplitude $a_{av}$. Data courtesy of prof. W. Szelenberger from the Medical University of Warsaw.

Spectral estimates (rows a and c) were calculated as follows: for each 4-seconds epoch, marked by expert as artifact-free and belonging to stage II for a and stage III or IV for c, power spectral density was calculated using single Hanning window of 512 points. Red curves present the average of these estimates in the relevant frequency bands. After integrating the spectral power in these bands for each epoch, average and mean square error (indicated on the right of each plot) were calculated. For both sleep-inducing drugs we observe an increase of power in the spectral band related to sleep spindles and decrease in SWA, in agreement with the physiological expectations. MP estimates of energy were calculated only for signal structures conforming to the generally accepted criteria for sleep spindles and SWA, based upon definitions used for standardization of visual scoring. Criteria for sleep spindles included: frequency 11-15 Hz², time duration 0.5-2.5 sec., peak-to-peak amplitude above 15 $\mu V$. Concordance of such selection of sleep spindles with visual detection was evaluated e.g. in [15]. Corresponding ranges of parameters for SWA were chosen as frequency 0.5-4 Hz, duration above 0.5 sec., amplitude higher than 75 $\mu V$. These structures are presented in rows b and c of Figure 3. Each of the blue dots in the frequency-amplitude coordinates marks one structure detected in one of the EEG epochs, selected as for the spectral estimates above. Numbers on the right of each plot indicate their average power, average number of structures per minute, total time of artifact-free recording from qualifying stages, average frequency, its variance and the average amplitude. We observe that power estimates calculated from structures selected from MP decomposition are significantly lower than the spectral power calculated in the same frequency band. MP estimates were calculated only for structures conforming to the relevant criteria, so energy carried by unrelated waveforms, occurring in the same frequency band, was not included. MP-based estimates proved to be more sensitive to the effects of drug influence on EEG than the spectral integrals. For the presented recordings the power of sleep spindles showed an average increase of 59% comparing to 26% increase of the spectral integrals within the same band. Average decrease of SWA power was 42%, comparing to 30% decrease of the corresponding spectral integral. Similar improvement was observed also in the other of 8 cases analyzed in this study. Apart from the higher sensitivity, such a detailed description allows for a closer investigation e.g. of the causes of changes in the power estimates. For example, we observe from the numerical values given in Figure 3, that the change of total power of both sleep spindles and SWA results from a change in the number of their occurrences per minute rather than change of the average structure's amplitude.

Footnotes

... Hz²

After formulation of the classical definition from [13], it was generally agreed that the frequency range 12-14Hz is too narrow for sleep spindles.