Generalizability of Frequency Weighting Curve for Extraction of Spectral Drowsy Component From the EEG Signals Recorded in Eyes-Closed Condition.

One of the critical barriers to reducing the threats of sleep loss to public health, safety, and productivity is a lack of practical tools for quick identification of objective level of sleepiness. We examined a novel sleepiness measure named "spectral drowsy component score" to provide evidence for generalizability of a frequency weighting curve required for calculation of this measure. Each spectral drowsy component score is a sum of 16 weighted ln-transformed single-Hz power densities (1-16 Hz) obtained by the fast Fourier transformation of an electroencephalographic signal recorded during the first minute after closing the eyes. A set of 16 weights (frequency weighting curve) is derived empirically. One type of such curve is a correlation spectrum. It consists of 16 coefficients of correlation of a group-averaged experimental time course of sleepiness with16 time courses of single-Hz power densities. Sleepiness is determined either subjectively (by self-scoring on the Karolinska Sleepiness Scale) or objectively (as sleep latency). Another type is a differential spectrum reflecting difference between 2 sets of 16 power densities obtained for either distant phases of sleep deprivation experiment or distinct alertness-sleepiness substates. Analysis of 3 datasets collected in sleep deprivation experiments with, in total, 160 participants showed that, despite differences in the protocols of these experiments and ages of their participants, the forms of frequency weighting curves always resembled one another. Such resemblance led to practical identity of scoring results. We concluded that spectral drowsy component scoring might be implemented into quick, simple, direct, transparent, and objective test of sleepiness.