EEG recurrence markers and sleep quality

doi:10.1016/j.jns.2013.04.019

Journal of the Neurological Sciences

Volume 331, Issues 1–2, 15 August 2013, Pages 26-30

https://doi.org/10.1016/j.jns.2013.04.019 Get rights and content

Abstract

Objectives

To show that EEG markers formed using the variable percent recurrence reliably quantified two related aspects of sleep quality, sleep depth and sleep fragmentation. As hypotheses, the depth marker would increase and the fragmentation marker decrease in patients where improved sleep quality occurred when assessed by polysomnography.

Methods

The patients (N = 20) had been diagnosed with obstructive sleep apnea during diagnostic polysomnography (dPSG), and had exhibited increased REM sleep (clinical indication of improved sleep quality) during subsequent polysomnography to titrate the pressure of a treatment device (cPSG). Percent recurrence was computed second-by-second from the EEG; sleep-depth and sleep-stability markers were obtained algorithmically. By assumption, the markers contained temporal information regarding the extent of deterministic (non-random) brain activity. Marker means were compared between the dPSG and the cPSG for NREM and REM sleep.

Results

Sleep depth was greater and sleep fragmentation was less during cPSG, as hypothesized (P < 0.05). The effects occurred during NREM and REM sleep, but were greater during NREM sleep (P < 0.05). At least one of the predicted changes occurred in 95% of the patients.

Conclusions

The factors generally regarded as responsible for subjective sleep quality were objectively quantified on the basis of dynamical changes in the EEG.

Introduction

Human sleep and associated events are assessed on the basis of rules applied to simultaneously recorded physiological signals [1]. Three stages (N1, N2, N3) and particular arousal events (abrupt changes) are identified from the electroencephalogram (EEG), and a fourth stage (REM) is identified from the coordinated behavior of several signals including the EEG. The N3 stage is commonly regarded as deep sleep. The depth of sleep together with the rate of arousal events are determinants of sleep quality [2]. Loss of sleep depth and/or increases in arousal events produce non-restorative sleep, and are associated with various sleep disorders including obstructive sleep apnea (OSA).

Present methods for measuring the depth of sleep are problematical [3], [4], [5], [6], [7]. Determining the intensity of stimuli needed to wake a subject has been used to quantify sleep depth [3], [4], but that method variably classified REM as the deepest level of sleep [5], intermittently deep [6], or as similar in depth to N1 and N2 sleep [6], depending on how the threshold was measured. Delta power is a marker for sleep depth during non-REM (NREM) sleep [7], but no equivalent marker exists for REM sleep. Similarly, additional refinement of scoring arousals is needed [8]. We recently showed that a recurrence marker computed by algorithmic analysis of the EEG stratified all sleep stages, increased progressively with NREM sleep-stage depth (N1 < N2 < N3), and characterized sleep fragmentation caused by arousal events [9].

REM rebound (an increase in percent of overnight sleep that is staged as REM) occurs during recovery from chronic stress, including restorative sleep following sleep deprivation [10] and initiation of treatment for OSA using continuous positive airway pressure (CPAP) [11], [12], [13], [14]. CPAP-associated REM rebound (CARR) is generally accepted to indicate deeper and less fragmented sleep [11], [12], [13], [14]. We therefore expected an increase in recurrence in CARR patients and a decrease in the variability of the recurrence, compared with the corresponding values determined prior to initiation of CPAP.

Our goal was to evaluate the capability of the EEG-based recurrence variable percent recurrence to quantify sleep depth and sleep fragmentation. The first aim was to show that a recurrence depth marker increased in patients who experienced CARR (increased sleep depth). The second aim was to show in the same patients that a recurrence fragmentation marker exhibited a decreased rate of change (decreased sleep fragmentation).

Section snippets

Subjects

We reviewed consecutive records of patients who underwent attended overnight diagnostic polysomnography (dPSG) that was positive for OSA (apnea–hypopnea index (AHI) ≥ 5 events/hr), and who subsequently underwent overnight CPAP-titration polysomnography (cPSG) during which CARR (clinical indicator of increased sleep depth and improved sleep quality) was observed. CARR was defined as an increase in REM as a percentage of total sleep time of at least 20%. This threshold was higher than that used

Results

The second-by-second variation of brain electrical activity as reflected in r(t) differed profoundly as a consequence of treatment with CPAP (Fig. 2). In the dPSG, r(t) typically varied over its entire range regardless of sleep stage (Fig. 2a). In the cPSG, however, r(t) was bounded with the highest mean value occurring in N3, lowest in wake, and intermediate in N2 and REM (Fig. 2b). The hypnograms in the dPSGs and cPSGs effectively were averages of the temporal changes in r(t).

After

Discussion

Sleep depth and sleep fragmentation are continuous, deterministic (non-random) features of the instantaneous state of brain electrical activity (brain states), but suitable methods for quantifying the features have not been developed. Recurrence analysis is well suited to the task, at least to the extent that the features are reflected in the EEG, which is a temporal output signal of the brain. Our aim was to show that improvements in sleep depth and sleep fragmentation that were established by

Conflict of interest statement

The authors have no conflicts of interest to report.

References (23)

D.J. Buysse et al.
The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research
Psychiatry Res
(1989)
S. Grahnstedt et al.
Awakening thresholds for electrical brain stimulation in five sleep-waking stages in the cat
Electroencephalogr Clin Neurophysiol
(1980)
D.J. Dijk et al.
All night spectral analysis of EEG sleep in young adult and middle-aged male subjects
Neurobiol Aging
(1989)
S. Carrubba et al.
Continuous EEG-based dynamic markers for sleep depth and phasic events
J Neurosci Methods
(2012)
L. Parrino et al.
Reorganization of sleep patterns in severe OSAS under prolonged CPAP treatment
Clin Neurophysiol
(2005)
A. Verma et al.
Slow wave sleep rebound and REM rebound following the first night of treatment with CPAP for sleep apnea: correlation with subjective improvement in sleep quality
Sleep Med
(2001)
S. Carrubba et al.
Method for detection of changes in the EEG induced by the presence of sensory stimuli
J Neurosci Methods
(2008)
American Academy of Sleep Medicine
The AASM manual for the scoring of sleep and associated events: rules, terminology and technical specifications: American Academy of Sleep Medicine
(2007)
A.A. Borbély et al.
Sleep homeostasis and models of sleep regulation
S.M. McKinney et al.
Covert waking brain activity reveals instantaneous sleep depth
PLoS One
(2011)

A. Rechtschaffen et al.

Auditory awakening thresholds in REM and NREM sleep stages

Percept Mot Skills

(1966)

Cited by (11)

An original method for staging sleep based on dynamical analysis of a single EEG signal
2018, Journal of Neuroscience Methods
Citation Excerpt :
Both variables quantified the complexity of the signal in the sense that higher values indicate less complexity (Zbilut and Webber, 2006). Both r and d were minimal during wake and maximal during deep sleep (Carrubba et al., 2012b; Wang et al., 2013), indicating that they could be interpreted as measures of sleep depth. The variables were computed second by second, resulting in chronological sequences (r(t), d(t)) that consisted of approximately 28,800 points for a typical 8-hr EEG.
The dynamical complexity of brain electrical activity manifested in the EEG is quantifiable using recurrence analysis (RA). Employing RA, we described and validated an originative method for automatically classifying epochs of sleep that is conceptually and instrumentally distinct from the existing method.
Complexity in single overnight EEGs was characterized second-by-second using four RA variables that were each averaged over consecutive 30-sec epochs to form four-component vectors. The vectors were staged using four-component cluster analysis. Method validity and utility were established by showing: (1) inter- and intra-subject consistency of staging results (method insusceptible to nonstationarity of the EEG); (2) use of method to eliminate costly and arduous visual staging in a binary classifications task for detecting a neurogenic disorder; (3) ability of method to provide new physiological insights into brain activity during sleep.
RA of sleep-acquired EEGs yielded four continuous measures of complexity and its change-rate that allowed automatic classification of epochs into four statistically distinct clusters (“stages”). Matched subjects with and without mental distress were accurately classified using biomarkers based on stage designations.
For binary-classification purposes, the method was cheaper, faster, and at least as accurate as the existing staging method. Epoch-by-epoch comparison of new versus existing methods revealed that the latter assigned epochs having widely different dynamical complexities into the same stage (dynamical incoherence).
Sleep can be automatically staged using an originative method that is fundamentally different from the existing method.
Reprint of “A new approach to analyze data from EEG-based concealed face recognition system”
2017, International Journal of Psychophysiology
The purpose of this study is to extend a feature set with non-linear features to improve classification rate of guilty and innocent subjects. Non-linear features can provide extra information about phase space. The Event-Related Potential (ERP) signals were recorded from 49 subjects who participated in concealed face recognition test. For feature extraction, at first, several morphological characteristics, frequency bands, and wavelet coefficients (we call them basic-features) are extracted from each single-trial ERP. Recurrence Quantification Analysis (RQA) measures are then computed as non-linear features from each single-trial. We apply Genetic Algorithm (GA) to select the best feature set and this feature set is used for classification of data using Linear Discriminant Analysis (LDA) classifier. Next, we use a new approach to improve classification results based on introducing an adaptive-threshold. Results indicate that our method is able to correctly detect 91.83% of subjects (45 correct detection of 49 subjects) using combination of basic and non-linear features, that is higher than 87.75% for basic and 79.59% for non-linear features. This shows that combination of non-linear and basic- features could improve classification rate.
A new approach to analyze data from EEG-based concealed face recognition system
2017, International Journal of Psychophysiology
Citation Excerpt :
Another non-linear analysis method is Recurrence Plots (RPs) that can be used even for short and non-stationary time series (Schinkel et al., 2007). RPs and Recurrence Quantification Analysis (RQA) have been successfully applied to several EEG processing studies, such as episodic memory retrieval (Talebi and Nasrabadi, 2010; Ghoshuni et al., 2007), ERP components detection (Schinkel et al., 2009; Marwan et al., 2007), epileptic seizure detection (Thomasson et al., 2010), sleep (Wang et al., 2013) and anesthesia analysis (Shalbaf et al., 2014). In this study we extracted RQA measures from single-trial ERPs and used them together with some morphological, frequency and wavelet features as feature set.
The purpose of this study is to extend a feature set with non-linear features to improve classification rate of guilty and innocent subjects. Non-linear features can provide extra information about phase space. The Event-Related Potential (ERP) signals were recorded from 49 subjects who participated in concealed face recognition test. For feature extraction, at first, several morphological characteristics, frequency bands, and wavelet coefficients (we call them basic-features) are extracted from each single-trial ERP. Recurrence Quantification Analysis (RQA) measures are then computed as non-linear features from each single-trial. We apply Genetic Algorithm (GA) to select the best feature set and this feature set is used for classification of data using Linear Discriminant Analysis (LDA) classifier. Next, we use a new approach to improve classification results based on introducing an adaptive-threshold. Results indicate that our method is able to correctly detect 91.83% of subjects (45 correct detection of 49 subjects) using combination of basic and non-linear features, that is higher than 87.75% for basic and 79.59% for non-linear features. This shows that combination of non-linear and basic- features could improve classification rate.
Clinical factors associated with extreme sleep apnoea [AHI > 100 events per hour]in peruvian patients: A case-control study–A preliminary report
2015, Sleep Science
Citation Excerpt :
This finding is relevant to the pathogenesis of OSA, since a high correlation has been identified between an increase in neck circumference and AHI value [27]. There is only one reference in the literature where this population of Extreme sleep apnoea is distinguished from the other subgroups of severe sleep apnoea [28], on this paper, the only patients who has significant REM rebound were the patients who has AHI>100 on the baseline portion of sleep study. In our population this occurrence was not observed.
The severity of obstructive sleep apnoea (OSA) ranges from mild or moderate to severe sleep apnoea. However, there is no information available on the clinical characteristics associated with cases involving more than 100 events per hour. This is a preliminary report and our goal was to characterise the demographics and sleep characteristics of patients with Extreme OSA and compare with patients with sleep apnoea of lesser severity. We hypothesised that patients with Extreme OSA (AHI>100) is associated with an increased comorbidities and/or risk factors.
We carried out a case-control study on male patients with OSA who were seen in a private hospital in Lima, Peru between 2006 and 2012. Cases were identified if their apnoea/hypopnea index (AHI) was higher than 100 (Extreme OSA), and four controls were selected per case: two with 15–29 AHI and two with 30–50 AHI, matched according to case diagnosis dates. We evaluated demographic, past medical history, and oxygen saturation variables
We identified 19 cases that were matched with 54 controls. In the multivariate model, only arterial hypertension, neck circumference, age, and over 10% in SatO2Hb≤90% in total sleep time (T90) were associated with Extreme OSA. Arterial hypertension had an OR=6.31 (CI95%: 1.71–23.23) of Extreme OSA. Each 5-cm increment in neck circumference was associated with an increase of OR=4.34 (CI95%: 1.32–14.33), while T90>10% had an OR=19.68 (CI95%: 4.33–89.49). Age had a marginal relevance (OR=0.95; CI95%: 0.92–0.99)
Our results suggest that arterial hypertension, neck circumference, and over 10% SatO2Hb≤90% in total sleep time were associated with a higher probability of Extreme OSA. We recommend investigators to study this population of Extreme OSA looking for an early diagnosis and the identification of prognostic factors in comparison with moderate to severe levels.
Two-group classification of patients with obstructive sleep apnea based on analysis of brain recurrence
2014, Clinical Neurophysiology
To demonstrate that the severity of obstructive sleep apnea (OSA) could be predicted algorithmically by means of recurrence analysis of the sleep-staged electroencephalogram (EEG).
A randomly selected cohort of 20 sleep-staged patients with OSA (apnea–hypopnea index (AHI) 5–30) was divided into mild and moderate sub-cohorts (AHI 5–15, 16–30, respectively), and the sleep EEG (C3) was analyzed using analysis of brain recurrence (ABR) (LSU cohort). Twenty distinct but related markers for sleep depth and fragmentation were computed from four ABR variables, and a marker function capable of classifying each patient into one of the two sub-cohorts was determined by linear discriminant analysis. Classification accuracy of individual patients was evaluated using area under the receiver operator characteristics curve (AUROC). As a control procedure, 20 additional sleep-staged patients with OSA whose polysomnographic data was obtained from an independent database were also evaluated (SHHS cohort).
On average, markers for sleep depth were reduced and those for sleep fragmentation were increased in the patients with moderate OSA, as expected. All patients in both cohorts were correctly classified using as few as 5–6 markers.
The degree of severity of OSA was reflected in objective changes in the sleep EEG. Recurrence analysis of the EEG potentially has uses beyond identification of the degree of OSA.
Recurrence analysis of the EEG during sleep accurately identifies subjects with mental health symptoms
2014, Psychiatry Research - Neuroimaging
Citation Excerpt :
Patients with multiple sclerosis were identified using ABR (Carrubba et al., 2010; Carrubba et al., 2012b), and it was used to create a novel paradigm in which the concepts of sleep depth and variability could be quantified (Carrubba et al., 2012a; Wang et al., 2013). Employing markers based on these variables, patients with mild or moderate obstructive sleep apnea were distinguished using the sleep-staged EEG from a single derivation (Wang et al., 2013), illustrating the concept that a complex physiologic disorder leaves an objectively discernible and specific footprint on brain electrical activity. We became interested in whether the sleep-acquired EEG could similarly be used to classify subjects with psychological distress.
Analysis of brain recurrence (ABR) is a novel computational method that uses two variables for sleep depth and two for sleep fragmentation to quantify temporal changes in non-random brain electrical activity. We postulated that ABR of the sleep-staged EEG could identify an EEG signature specific for the presence of mental health symptoms. Using the Mental Health Inventory Questionnaire (MHI-5) as ground truth, psychological distress was assessed in a study cohort obtained from the Sleep Heart Health Study. Subjects with MHI-5 <50 (N=34) were matched for sex, BMI, age, and race with 34 subjects who had MHI-5 scores >50. Sixteen ABR markers derived from the EEG were analyzed using linear discriminant analysis to identify marker combinations that reliably classified individual subjects. A biomarker function computed from 12 of the markers accurately classified the subjects based on their MHI-5 scores (AUROC=82%). Use of additional markers did not improve classification accuracy. Subgroup analysis (20 highest and 20 lowest MHI-5 scores) improved classification accuracy (AUROC=89%). Biomarker values for individual subjects were significantly correlated with MHI-5 score (r=0.36, 0.54 for N=68, 40, respectively). ABR of EEGs obtained during sleep successfully classified subjects with regard to the severity of mental health symptoms, indicating that mood systems were reflected in brain electrical activity.

View all citing articles on Scopus

View full text

EEG recurrence markers and sleep quality

Abstract

Objectives

Methods

Results

Conclusions

Introduction

Section snippets

Subjects

Results

Discussion

Conflict of interest statement

Psychiatry Res

Electroencephalogr Clin Neurophysiol

Neurobiol Aging

J Neurosci Methods

Clin Neurophysiol

Sleep Med

J Neurosci Methods

The AASM manual for the scoring of sleep and associated events: rules, terminology and technical specifications: American Academy of Sleep Medicine

Sleep homeostasis and models of sleep regulation

Covert waking brain activity reveals instantaneous sleep depth

PLoS One

Auditory awakening thresholds in REM and NREM sleep stages

Percept Mot Skills