A Quantitative Framework for Assessing Sleep Quality from EEG Time Series in Complex Dynamic Systems

TL;DR

Sleep quality (SQ) is quantified objectively by examining EEG phase-amplitude coupling (PAC), particularly delta-beta PAC, across sleep and waking states including resting-state. The study integrates spectrogram, weighted phase lag index (wPLI), and PAC analyses with a four-session within-subject design and leave-one-subject-out cross-validation to classify SQ at the individual level, finding delta-beta PAC to outperform other EEG features. Good sleepers exhibit stronger delta-beta PAC, which correlates with SQ as measured by subjective indices, supporting PAC as a robust biomarker for SQ and its neural determinants. The work advances objective sleep health assessment and provides mechanistic insights with potential clinical applications for sleep disorders and cognitive health.

Abstract

Modern lifestyles contribute to insufficient sleep, impairing cognitive function and weakening the immune system. Sleep quality (SQ) is vital for physiological and mental health, making its understanding and accurate assessment critical. However, its multifaceted nature, shaped by neurological and environmental factors, makes precise quantification challenging. Here, we address this challenge by utilizing electroencephalography (EEG) for phase-amplitude coupling (PAC) analysis to elucidate the neurological basis of SQ, examining both states of sleep and wakefulness, including resting state (RS) and working memory. Our results revealed distinct patterns in beta power and delta connectivity in sleep and RS, together with the reaction time of working memory. A notable finding was the pronounced delta-beta PAC, a feature markedly stronger in individuals with good SQ. We further observed that SQ was positively correlated with increased delta-beta PAC. Leveraging these insights, we applied machine learning models to classify SQ at an individual level, demonstrating that the delta-beta PAC outperformed other EEG characteristics. These findings establish delta-beta PAC as a robust electrophysiological marker to quantify SQ and elucidate its neurological determinants.

Figures (10)

• Figure 1: Experimental setup. (a) The experiment consists of four main sessions: pre-nap, a 90-minute nap, post-nap, and post-night. The questionnaire (Q) component, comprising the Stanford Sleepiness Scale and the Brunel Mood Scale, is conducted alongside wakefulness, including working memory tasks and 5-minute eyes-closed resting state (RS). The RSs comprise pre-nap (RS 1 to RS 4), post-nap (RS 5 to RS 6), and post-night (RS 7 to RS 8). (b) Working memory tasks are composed of encoding (E) and retrieval (R) phases. The word-pair (WP) memory task consists of 54 trials, while the visuospatial (VS) memory task includes 38 trials for the encoding phase and 76 trials for the retrieval phase. (c) Electrode placement for the 60-channel EEG and 4-channel EOG is grouped into five brain regions: frontal, central, temporal, parietal, and occipital.
• Figure 2: An overall framework for classifying sleep quality based on EEG features by sessions: pre-nap, nap, post-nap, and post-night.
• Figure 3: Demographics and sleep architecture between good and poor sleepers (lower quartile-upper quartile).
• Figure 4: Performance on (a) accuracy and (b) reaction time in three memory tasks for good and poor sleepers across the pre-nap, post-nap, and post-night sessions. A black asterisk indicates statistical significance (p$<$ 0.05, FDR corrected).
• Figure 5: Statistical t-value maps showing the relative increase and decrease in spectrogram between two groups across (a) sleep stages and (b) wakefulness. This analysis highlights the frequency differences in the range of 0.5-30 Hz. Channels corresponding to the regions of interest (ROIs) – frontal, central, temporal, parietal, and occipital – are averaged. The black dotted boxes and white asterisks indicate significant areas (p$<$ 0.05, FDR corrected).