EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

The Hypno-PC: uncovering sleep dynamics through principal component analysis and hidden Markov modeling of electrophysiological signals

Miriam Guendelman iD, Oren Shriki iD

SLEEPJ · 2026

DOI: 10.1093/sleep/zsaf164 ↗

Kurz gefasst

Fachartikel in der Zeitschrift SLEEPJ, 2026. Verfasst von Miriam Guendelman und 1 weiteren Autor:innen. Beteiligte Einrichtung: Department of Cognitive and Brain Sciences, Ben-Gurion University of the Negev, Beer-Sheva, Israel. Hauptthemen laut MEDLINE: Principal Component Analysis, Sleep Stages, Sleep. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 61 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: Israel Science Foundation.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Manual sleep scoring segments sleep into discrete 30-s epochs (wake, non-rapid-eye-movement [NREM] 1–3, rapid-eye-movement [REM]), yet substantial evidence suggests that sleep unfolds as a continuous, microstate-rich process. Using a data-driven approach, we analyzed overnight high-density electroencephalography, electrooculography, electromyography, and electrocardiography recordings from 29 healthy adults (ANPHY-Sleep dataset). Signal-specific features from standard 30-s and finer 4-s epochs were compressed using principal component analysis (PCA). With the 30-s epochs, the first principal component (PC) (Hypno-PC; 42 per cent variance) closely tracked the hypnogram, while the extended PCA space (explaining 90 per cent variance) achieved sleep-stage separability comparable to the state-of-the-art YASA classifier. Furthermore, Hypno-PC emphasized continuous sleep dynamics, revealing a gradual descent into deep NREM sleep contrasted with abrupt transitions into REM or wakefulness. Independent component analysis (ICA) on the top PCs (n = 5) separated spindle-rich, slow-wave-dominant, and arousal-related processes. A Gaussian hidden Markov model (GHMM) fitted to ICA features identified four macrostates at 30-s resolution, aligning closely with canonical sleep stages (Kappa = 0.70). These macrostates required minimal labeling (<1% of epochs) and provided highly accurate estimates of sleep-onset latency. At a finer 4-s resolution, the GHMM resolved eleven microstates, distinguishing tonic from phasic REM, active from quiet wakefulness, and early- from late-night NREM subtypes. Three hub states—active wake, N1-like, and late slow-wave-rich—mediated most microstate transitions, highlighting structured continuity within sleep microstate architecture. This linear, interpretable PCA–ICA–GHMM framework bridges conventional sleep staging, continuous sleep dynamics, and detailed microstate structure, offering clinicians and researchers a scalable, objective tool for studying sleep architecture. Statement of Significance This study introduces a data-driven framework that bridges traditional sleep scoring and the intrinsic continuity of human sleep. Using high-density electroencephalography, electrooculography, electromyography, and electrocardiography, we derive a low-dimensional space that captures sleep architecture through unsupervised methods. The leading dimension, explaining most of the signal’s variability, faithfully tracks the hypnogram while revealing gradual descents into deep non-rapid-eye-movement (NREM) and abrupt shifts into rapid-eye-movement (REM) or brief awakenings. We further reveal spindle-rich, slow-wave, and arousal components and identify data-driven states closely aligning with canonical sleep stages. At finer temporal resolution, we uncover structured microstates, distinguishing tonic versus phasic REM and early versus late NREM. Our interpretable principal component analysis–independent component analysis–Gaussian hidden Markov model pipeline thus provides clinicians and researchers with a scalable tool for objective sleep phenotyping.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Principal Component AnalysisSleep StagesSleepHumansMarkov ChainsElectroencephalographyAdultElectromyographyElectrooculographyMaleFemaleElectrocardiographyPolysomnographyYoung Adult

Stichwörter der Autor:innen

Gaussian hidden Markov model (GHMM)automated polysomnography analysiscontinuous sleep processeshigh-density EEGindependent component analysis (ICA)principal component analysis (PCA)sleep microstatesunsupervised machine learning in sleep

Autor:innen und Institutionen

  • Department of Cognitive and Brain Sciences, Ben-Gurion University of the Negev, Beer-Sheva, Israel.
  • Department of Psychology, Ben-Gurion University of the Negev, Beer-Sheva, Israel.

ORCID-Profile: Miriam Guendelman, Oren Shriki

Förderung

  • Israel Science Foundation

Zitieren

APA

Guendelman, M., & Shriki, O. (2026). The Hypno-PC: uncovering sleep dynamics through principal component analysis and hidden Markov modeling of electrophysiological signals. SLEEPJ. https://doi.org/10.1093/sleep/zsaf164

Vancouver

Guendelman M, Shriki O. The Hypno-PC: uncovering sleep dynamics through principal component analysis and hidden Markov modeling of electrophysiological signals. SLEEPJ. 2026. doi:10.1093/sleep/zsaf164

BibTeX
@article{Guendelman2026,
  author = {Guendelman, Miriam and Shriki, Oren},
  title = {The Hypno-PC: uncovering sleep dynamics through principal component analysis and hidden Markov modeling of electrophysiological signals},
  journal = {SLEEPJ},
  year = {2026},
  doi = {10.1093/sleep/zsaf164}
}
Kostenlos starten →