Benchmarking performance of an automatic polysomnography scoring system in a population with suspected sleep disorders.

Aim: The current gold standard for measuring sleep disorders is polysomnography (PSG), which is manually scored by a sleep technologist. Scoring a PSG is time-consuming and tedious, with substantial inter-rater variability. A deep-learning-based sleep analysis software module can perform autoscoring of PSG. The primary objective of the study is to validate the accuracy and reliability of the autoscoring software. The secondary objective is to measure workflow improvements in terms of time and cost via a time motion study. Methodology: The performance of an automatic PSG scoring software was benchmarked against the performance of two independent sleep technologists on PSG data collected from patients with suspected sleep disorders. The technologists at the hospital clinic and a third-party scoring company scored the PSG records independently. The scores were then compared between the technologists and the automatic scoring system. An observational study was also performed where the time taken for sleep technologists at the hospital clinic to manually score PSGs was tracked, along with the time taken by the automatic scoring software to assess for potential time savings. Results: Pearson's correlation between the manually scored apnea-hypopnea index (AHI) and the automatically scored AHI was 0.962, demonstrating a near-perfect agreement. The autoscoring system demonstrated similar results in sleep staging. The agreement between automatic staging and manual scoring was higher in terms of accuracy and Cohen's kappa than the agreement between experts. The autoscoring system took an average of 42.7 s to score each record compared with 4,243 s for manual scoring. Following a manual review of the auto scores, an average time savings of 38.6 min per PSG was observed, amounting to 0.25 full-time equivalent (FTE) savings per year. Conclusion: The findings indicate a potential for a reduction in the burden of manual scoring of PSGs by sleep technologists and may be of operational significance for sleep laboratories in the healthcare setting.

An automated heart rate-based algorithm for sleep stage classification: Validation using conventional polysomnography and an innovative wearable electrocardiogram device.

Background: The rapid advancement in wearable solutions to monitor and score sleep staging has enabled monitoring outside of the conventional clinical settings. However, most of the devices and algorithms lack extensive and independent validation, a fundamental step to ensure robustness, stability, and replicability of the results beyond the training and testing phases. These systems are thought not to be feasible and reliable alternatives to the gold standard, polysomnography (PSG). Materials and methods: This validation study highlights the accuracy and precision of the proposed heart rate (HR)-based deep-learning algorithm for sleep staging. The illustrated solution can perform classification at 2-levels (Wake; Sleep), 3-levels (Wake; NREM; REM) or 4- levels (Wake; Light; Deep; REM) in 30-s epochs. The algorithm was validated using an open-source dataset of PSG recordings (Physionet CinC dataset, n = 994 participants, 994 recordings) and a proprietary dataset of ECG recordings (Z3Pulse, n = 52 participants, 112 recordings) collected with a chest-worn, wireless sensor and simultaneous PSG collection using SOMNOtouch. Results: We evaluated the performance of the models in both datasets in terms of Accuracy (A), Cohen's kappa (K), Sensitivity (SE), Specificity (SP), Positive Predictive Value (PPV), and Negative Predicted Value (NPV). In the CinC dataset, the highest value of accuracy was achieved by the 2-levels model (0.8797), while the 3-levels model obtained the best value of K (0.6025). The 4-levels model obtained the lowest SE (0.3812) and the highest SP (0.9744) for the classification of Deep sleep segments. AHI and biological sex did not affect scoring, while a significant decrease of performance by age was reported across the models. In the Z3Pulse dataset, the highest value of accuracy was achieved by the 2-levels model (0.8812), whereas the 3-levels model obtained the best value of K (0.611). For classification of the sleep states, the lowest SE (0.6163) and the highest SP (0.9606) were obtained for the classification of Deep sleep segment. Conclusion: The results of the validation procedure demonstrated the feasibility of accurate HR-based sleep staging. The combination of the proposed sleep staging algorithm with an inexpensive HR device, provides a cost-effective and non-invasive solution deployable in the home environment and robust across age, sex, and AHI scores.

Vigilance declines following sleep deprivation are associated with two previously identified dynamic connectivity states.

Robustly linking dynamic functional connectivity (DFC) states to behaviour is important for establishing the utility of the method as a functional measurement. We previously used a sliding window approach to identify two dynamic connectivity states (DCS) related to vigilance. A new sample of 32 healthy participants underwent two sets of task-free functional magnetic resonance imaging (fMRI) scans, once in a well-rested state and once after a single night of total sleep deprivation. Using a temporal difference method, DFC and clustering analysis on the task-free fMRI data revealed five centroids that were highly correlated with those found in previous work. In particular, two of these states were associated with high and low arousal respectively. Individual differences in vulnerability to sleep deprivation were measured by assessing state-related changes in Psychomotor Vigilance Test (PVT) performance. Changes in the duration spent in each of the arousal states from the well-rested to the sleep-deprived condition correlated with declines in PVT performance. The reproducibility of DFC measures and their association with vigilance highlight their utility in serving as a neuroimaging method with behavioural relevance. (178 words).

Large-scale data from wearables reveal regional disparities in sleep patterns that persist across age and sex.

Prior reports on geographical differences in sleep duration have relied on samples collected at different time points with a variety of subjective instruments. Using sleep data from a total of 553,559 nights from 23,680 Fitbit users (aged 15-80y), we found objective evidence for regional disparities in sleep duration of 32-43 min between Oceanian and East Asian users on weekdays. This was primarily driven by later bedtimes in East Asians. Although users in all countries extended sleep on weekends, East Asians continued to sleep less than their Oceanian counterparts. Women generally slept more than men, and older users slept less than younger users. Reasons for shorter sleep duration in East Asians on both weekdays and weekends, across the lifespan and in both sexes remain to be investigated.

Trait-like characteristics of sleep EEG power spectra in adolescents across sleep opportunity manipulations.

The electroencephalographic power spectra of non-rapid eye movement sleep in adults demonstrate trait-like consistency within participants across multiple nights, even when prior sleep deprivation is present. Here, we examined the extent to which this finding applies to adolescents who are habitually sleep restricted on school-days and sleep longer on weekends. We evaluated 78 adolescents across three sleep restriction groups who underwent different permutations of adequate sleep (9 hr time-in-bed), sleep restriction (5 hr time-in-bed), afternoon naps (1 hr afternoon) and recovery sleep (9 hr time-in-bed) that simulate behaviour on school-days and weekends. The control group comprised a further 22 adolescents who had 9 hr of sleep opportunity each night. Intra-class correlation coefficients showed moderate to almost perfect within-subject stability in electroencephalographic power spectra across multiple nights in both sleep restriction and control groups, even when changes to sleep macrostructure were observed. While nocturnal intra-class correlation metrics were lower in the low-frequency and spindle frequency bins in the sleep restriction compared with the control group, hierarchical clustering measures could still identify multi-night electroencephalographic spectra as originating from the same individual. The trait-like characteristics of electroencephalographic spectra from an adolescent remain identifiable despite the disruptive effects of multi-night sleep restriction to sleep architecture.

Dynamic functional connectivity markers of objective trait mindfulness.

While mindfulness is commonly viewed as a skill to be cultivated through practice, untrained individuals can also vary widely in dispositional mindfulness. Prior research has identified static neural connectivity correlates of this trait. Here, we use dynamic functional connectivity (DFC) analysis of resting-state fMRI to study time-varying connectivity patterns associated with naturally varying and objectively measured trait mindfulness. Participants were selected from the top and bottom tertiles of performers on a breath-counting task to form high trait mindfulness (HTM; N = 21) and low trait mindfulness (LTM; N = 18) groups. DFC analysis of resting state fMRI data revealed that the HTM group spent significantly more time in a brain state associated with task-readiness - a state characterized by high within-network connectivity and greater anti-correlations between task-positive networks and the default-mode network (DMN). The HTM group transitioned between brain states more frequently, but the dwell time in each episode of the task-ready state was equivalent between groups. These results persisted even after controlling for vigilance. Across individuals, certain connectivity metrics were weakly correlated with self-reported mindfulness as measured by the Five Facet Mindfulness Questionnaire, though these did not survive multiple comparisons correction. In the static connectivity maps, HTM individuals had greater within-network connectivity in the DMN and the salience network, and greater anti-correlations between the DMN and task-positive networks. In sum, DFC features robustly distinguish HTM and LTM individuals, and may be useful biological markers for the measurement of dispositional mindfulness.

Dynamic functional connectivity and its behavioral correlates beyond vigilance.

Fluctuations in resting-state functional connectivity and global signal have been found to correspond with vigilance fluctuations, but their associations with other behavioral measures are unclear. We evaluated 52 healthy adolescents after a week of adequate sleep followed by five nights of sleep restriction to unmask inter-individual differences in cognition and mood. Resting state scans obtained at baseline only, analyzed using sliding window analysis, consistently yielded two polar dynamic functional connectivity states (DCSs) corresponding to previously reported 'low arousal' and 'high arousal' states. We found that the relative temporal preponderance of two dynamic connectivity states (DCS) in well-rested participants, indexed by a median split of participants, based on the relative time spent in these DCS, revealed highly significant group differences in vigilance at baseline and its decline following multiple nights of sleep restriction. Group differences in processing speed and working memory following manipulation but not at baseline suggest utility of DCS in predicting cognitive vulnerabilities unmasked by a stressor like sleep restriction. DCS temporal predominance was uninformative about mood and sleepiness speaking to specificity in its behavioral predictions. Global signal fluctuation provided information confined to vigilance. This appears to be related to head motion, which increases during periods of low arousal.

An end-to-end framework for real-time automatic sleep stage classification.

Sleep staging is a fundamental but time consuming process in any sleep laboratory. To greatly speed up sleep staging without compromising accuracy, we developed a novel framework for performing real-time automatic sleep stage classification. The client-server architecture adopted here provides an end-to-end solution for anonymizing and efficiently transporting polysomnography data from the client to the server and for receiving sleep stages in an interoperable fashion. The framework intelligently partitions the sleep staging task between the client and server in a way that multiple low-end clients can work with one server, and can be deployed both locally as well as over the cloud. The framework was tested on four datasets comprising ≈1700 polysomnography records (≈12000 hr of recordings) collected from adolescents, young, and old adults, involving healthy persons as well as those with medical conditions. We used two independent validation datasets: one comprising patients from a sleep disorders clinic and the other incorporating patients with Parkinson's disease. Using this system, an entire night's sleep was staged with an accuracy on par with expert human scorers but much faster (≈5 s compared with 30-60 min). To illustrate the utility of such real-time sleep staging, we used it to facilitate the automatic delivery of acoustic stimuli at targeted phase of slow-sleep oscillations to enhance slow-wave sleep.

Auditory stimulation of sleep slow oscillations modulates subsequent memory encoding through altered hippocampal function.

Study Objectives: Slow oscillations (SO) during sleep contribute to the consolidation of learned material. How the encoding of declarative memories during subsequent wakefulness might benefit from their enhancement during sleep is less clear. In this study, we investigated the impact of acoustically enhanced SO during a nap on subsequent encoding of declarative material. Methods: Thirty-seven healthy young adults were studied under two conditions: stimulation (STIM) and no stimulation (SHAM), in counter-balanced order following a night of sleep restriction (4 hr time-in-bed [TIB]). In the STIM condition, auditory tones were phase-locked to the SO up-state during a 90 min nap opportunity. In the SHAM condition, corresponding time points were marked but tones were not presented. Thirty minutes after awakening, participants encoded pictures while undergoing fMRI. Picture recognition was tested 60 min later. Results: Acoustic stimulation augmented SO across the group, but there was no group level benefit on memory. However, the magnitude of SO enhancement correlated with greater recollection. SO enhancement was also positively correlated with hippocampal activation at encoding. Although spindle activity increased, this did not correlate with memory benefit or shift in hippocampal signal. Conclusions: Acoustic stimulation during a nap can benefit encoding of declarative memories. Hippocampal activation positively correlated with SO augmentation.

Spontaneous eyelid closures link vigilance fluctuation with fMRI dynamic connectivity states.

Fluctuations in resting-state functional connectivity occur but their behavioral significance remains unclear, largely because correlating behavioral state with dynamic functional connectivity states (DCS) engages probes that disrupt the very behavioral state we seek to observe. Observing spontaneous eyelid closures following sleep deprivation permits nonintrusive arousal monitoring. During periods of low arousal dominated by eyelid closures, sliding-window correlation analysis uncovered a DCS associated with reduced within-network functional connectivity of default mode and dorsal/ventral attention networks, as well as reduced anticorrelation between these networks. Conversely, during periods when participants' eyelids were wide open, a second DCS was associated with less decoupling between the visual network and higher-order cognitive networks that included dorsal/ventral attention and default mode networks. In subcortical structures, eyelid closures were associated with increased connectivity between the striatum and thalamus with the ventral attention network, and greater anticorrelation with the dorsal attention network. When applied to task-based fMRI data, these two DCS predicted interindividual differences in frequency of behavioral lapsing and intraindividual temporal fluctuations in response speed. These findings with participants who underwent a night of total sleep deprivation were replicated in an independent dataset involving partially sleep-deprived participants. Fluctuations in functional connectivity thus appear to be clearly associated with changes in arousal.

Classifying vulnerability to sleep deprivation using baseline measures of psychomotor vigilance.

OBJECTIVES: To identify measures derived from baseline psychomotor vigilance task (PVT) performance that can reliably predict vulnerability to sleep deprivation. DESIGN: Subjects underwent total sleep deprivation and completed a 10-min PVT every 1-2 h in a controlled laboratory setting. Participants were categorized as vulnerable or resistant to sleep deprivation, based on a median split of lapses that occurred following sleep deprivation. Standard reaction time, drift diffusion model (DDM), and wavelet metrics were derived from PVT response times collected at baseline. A support vector machine model that incorporated maximum relevance and minimum redundancy feature selection and wrapper-based heuristics was used to classify subjects as vulnerable or resistant using rested data. SETTING: Two academic sleep laboratories. PARTICIPANTS: Independent samples of 135 (69 women, age 18 to 25 y), and 45 (3 women, age 22 to 32 y) healthy adults. INTERVENTIONS: In both datasets, DDM measures, number of consecutive reaction times that differ by more than 250 ms, and two wavelet features were selected by the model as features predictive of vulnerability to sleep deprivation. Using the best set of features selected in each dataset, classification accuracy was 77% and 82% using fivefold stratified cross-validation, respectively. MEASUREMENTS AND RESULTS: In both datasets, DDM measures, number of consecutive reaction times that differ by more than 250 ms, and two wavelet features were selected by the model as features predictive of vulnerability to sleep deprivation. Using the best set of features selected in each dataset, classification accuracy was 77% and 82% using fivefold stratified cross-validation, respectively. CONCLUSIONS: Despite differences in experimental conditions across studies, drift diffusion model parameters associated reliably with individual differences in performance during total sleep deprivation. These results demonstrate the utility of drift diffusion modeling of baseline performance in estimating vulnerability to psychomotor vigilance decline following sleep deprivation.

Predicting vulnerability to sleep deprivation using diffusion model parameters.

We used diffusion modelling to predict vulnerability to decline in psychomotor vigilance task (PVT) performance following a night of total sleep deprivation (SD). A total of 135 healthy young adults (69 women, age = 21.9 ± 1.7 years) participated in several within-subject cross-over design studies that incorporated the PVT. Participants were classified as vulnerable (lower tertile) or non-vulnerable (upper tertile) according to their change in lapse rate [lapse = reaction time (RT) ≥ 500 ms] between the evening before (ESD) and the morning after SD. RT data were fitted using Ratcliff's diffusion model. Although both groups showed significant change in RT during SD, there was no significant group difference in RT during the ESD session. In contrast, during ESD, the mean diffusion drift of vulnerable subjects was significantly lower than for non-vulnerable subjects. Mean drift and non-decision times were both adversely affected by sleep deprivation. Both mean drift and non-decision time showed significant state × vulnerability interaction. Diffusion modelling appears to have promise in predicting vulnerability to vigilance decline induced by a night of total sleep deprivation.