Crude EEG parameter provides sleep medicine with well-defined continuous hypnograms

TL;DR

This paper introduces a simple EEG parameter based on local maxima and minima that produces continuous hypnograms closely matching expert sleep stage annotations, offering a potential standardized biomarker for brain activity.

Contribution

The study presents a novel, straightforward algorithm for continuous hypnogram generation from EEG data, improving sleep analysis and revealing detailed sleep dynamics and oscillations.

Findings

Continuous hypnograms closely match expert annotations

Sleep depth decreases from evening to morning within stages

Infra-slow oscillations are prominent at sleep onset and in S2 phases

Abstract

To evaluate EEG data, one can count local maxima and minima on a fine scale, in a sliding window analysis. This straightforward calculation, which simplifies and improves previous work on permutation entropy, directly defines a good proxy for brain activity in an EEG channel during an epoch of 30 seconds. Different channels and persons can be compared when they are measured with the same device and prefiltering options. This could lead to a rigorously defined and suitably standardized biomarker of cortex activity, like blood pressure or laboratory values. Applied to sleep EEG, the algorithm yields hypnograms with continuous scale which show amazing coincidence with sleep stage annotation by trained experts. Although produced by a crude method, continuous hypnograms provide a lot of details. For example, sleep depth usually decreases from evening to morning even within the same annotated sleep stage, except for REM phases where mean sleep depth is rather constant, but different in frontal and parietal channels. The diagnostic potential of the method is demonstrated with two hypnograms of narcoleptic patients. In all 10 subjects, infra-slow oscillations of activity with a wavelength between 30s and two minutes were clearly seen, particularly strong at the onset of sleep and in S2 phases. The suggested method needs to be checked and improved. In its present form it seems already an appropriate tool for screening long-term EEG data.