Supervised learning-based automatic sleep stage classification methods rely on the availability of a large amount of labeled sleep data. On the other hand, in-home EEG devices are emerging as an alternative to polysomnography (PSG). A recent work employed contrastive learning - a self-supervised learning method - for label-efficient sleep stage classification on an in-home EEG dataset. However, this method classified each sleep epoch independently without considering the short-range temporal context. This work proposes FC-CL, a simple two-step framework that integrates temporal context at the feature level for contrastively pretrained sleep stage classification. Step 1 (CL) performs contrastive pretraining on unlabeled multi-channel in-home EEG, followed by supervised fine-tuning to obtain a baseline 5-stage classifier. Step 2 (FC) concatenates the per-epoch feature of a target epoch with those of its temporally surrounding epochs, and trains a classifier while keeping the pretrained encoder fixed. Thus, our method accounts for the transitional properties of the sleep stages without introducing complex sequential models. In the experiments, we demonstrate the effectiveness of FC-CL compared to other methods.

BACKGROUND:Cheyne-Stokes respiration (CSR) is a major form of central sleep apnea in patients with heart failure and is associated with mortality in these patients. Although full polysomnography is required for CSR diagnosis, access may be limited. Cyclic variation of heart rate (CVHR) detected by Holter electrocardiogram reflects changes in cardiac autonomic activity associated with apnea-hypopnea events and is easier to perform than polysomnography. METHODS:We examined whether RR interval shortening time in the CVHR analysis could be useful for screening CSR. In this study, 41 patients were analyzed. Holter electrocardiogram and polysomnography were simultaneously performed. CVHR events were classified as CSR or not based on RR interval shortening time, and the percentage of CSR (%CSR), defined as the ratio of total CSR duration to time in bed, was measured. RESULTS:There were 27 patients with heart failure. The patients tended to have severe sleep-disordered breathing and an apnea-hypopnea index of 38.9 (24.7), and seven patients had predominantly central events. Based on %CSR determined by polysomnography with manual scoring ≧20%, 10 patients were labeled as CSR-positive and 31 as CSR-negative. There was moderate-to-good reliability between the percentage of CSR determined via CVHR analysis and the percentage of CSR determined via polysomnography [intraclass correlation coefficient = 0.74 (0.56, 0.85); p < 0.001]. CONCLUSIONS:Holter electrocardiograms are performed for many patients with heart failure and are better tolerated than overnight polysomnography. Our method of using CVHR detected by Holter electrocardiogram could be helpful for CSR screening.

BACKGROUND:Quantitative thresholds for REM sleep without atonia (RWA) vary across populations, and their relationships with neuroimaging markers of Lewy body disease remain incompletely understood. OBJECTIVES:To examine whether previously proposed RWA thresholds for Japanese patients are associated with cardiac sympathetic and striatal dopaminergic imaging markers. METHODS:This multicenter study included 91 Japanese patients with probable RBD (isolated or PD-associated). Assessments included the RBD Questionnaire-Japanese version (RBDQ-JP), video-PSG, cardiac 123I-MIBG scintigraphy, and Dopamine transporter (DAT) imaging (123I-FP-CIT-SPECT). RWA was quantified using three methods: (1) the Sleep Innsbruck Barcelona (SINBAR) criteria, combining any submental activity and phasic activity in the flexor digitorum superficialis (FDS); (2) the AASM tonic criteria, assessing submental tonic activity; and (3) the AASM phasic criteria, assessing FDS phasic activity. RESULTS:97.8% (45/46) of patients meeting either the SINBAR ≥11.3% or AASM phasic ≥9.4% RWA threshold exhibited reduced cardiac MIBG uptake (H/M ratio <2.2), although the SINBAR threshold was substantially lower than the Western standard of 27.2%. No significant correlation existed between the specific binding ratio (SBR) and either %RWA or MDS-UPDRS-III scores. Notably, RWA and RBDQ-JP scores were only weakly correlated (rs = 0.14-0.37); among patients exceeding the RWA threshold, 19.7% scored below the RBDQ-JP cutoff of 19.5. CONCLUSIONS:The Japanese SINBAR and AASM phasic RWA thresholds are strongly associated with reduced cardiac MIBG uptake, but not with striatal DAT binding. These findings support an association between elevated RWA and cardiac sympathetic denervation, while suggesting that optimal thresholds vary across populations.

Obstructive sleep apnea (OSA) is closely associated with obesity and fluid retention. Our previous study suggested that, in patients with heart failure, the sodium-glucose cotransporter 2 inhibitor, tofogliflozin, promotes diuresis and weight loss and improves OSA severity. However, whether changes in the apnea-hypopnea index (AHI) are chronologically associated with body composition parameters remains unclear. We enrolled 10 patients (six men) with OSA. They received tofogliflozin (20 mg) daily for 6 months. The AHI was assessed using WatchPAT® at baseline, 3, and 6 months. Body composition, including water content and fat mass, was also measured. AHI changes measured using the WatchPAT peripheral arterial tonometry-derived AHI (pAHI) and their correlations with changes in body composition parameters were analyzed. Tofogliflozin administration significantly reduced pAHIs at 3 and 6 months compared to baseline, but more at 3 months, whereas body weight and body water content decreased over time. The

Purpose:We hypothesized that patients with head and neck cancer (HNC) have impaired pharynglaryngeal function during sleep compared with individuals without HNC, making them more susceptible to aspiration. We assumed polysomnography (PSG) would reveal decreased swallowing frequency, prolonged swallowing duration, reduced submental electromyographic (EMG) amplitude during swallowing, and abnormal timing of swallowing relative to respiratory phases. Patients and Methods:Patients with HNC who underwent radiotherapy to the pharyngolaryngeal region were enrolled. Age-, sex-, and body mass index-matched individuals without HNC served as controls. Swallowing frequency, duration, submental EMG amplitude during swallowing, and respiratory phases before and after swallowing were assessed using PSG and compared between groups. Data normality was assessed using the Shapiro-Wilk test, appropriate parametric or nonparametric analyses were applied with statistical significance set at p < 0.05. Results:Fifteen male patients with HNC and matched controls were analyzed. Swallowing frequency during sleep was significantly higher in the HNC group overall (p = 0.016), during rapid eye movement (REM) sleep (p = 0.033), and during non-REM sleep (p = 0.015). Swallowing duration during wakefulness was longer in the HNC group (p = 0.015), while no significant difference was observed during sleep (p = 0.73). Submental EMG amplitude during swallowing significantly decreased during sleep in the HNC group (p = 0.025). Respiratory pauses before and after swallowing were more frequent in the HNC group (p = 0.002 and p = 0.013, respectively). Conclusion:Swallowing function during sleep may be impaired in patients with HNC. Because the pharyngolaryngeal muscles are common to both swallowing and airway patency, dysfunction in this region may contribute to OSA-related breathing disturbances. Early identification and management of OSA, combined with targeted swallowing rehabilitation, may help reduce the risk of aspiration, improve airway safety, and potentially lower the risk of life-threatening complications, including sudden death.

OBJECTIVE:Hypoglossal nerve stimulation (HNS) has emerged as an effective treatment for obstructive sleep apnea (OSA) in patients who cannot tolerate continuous positive airway pressure. However, data on Asian populations, particularly Japanese patients, are limited. This study evaluated the operative time and clinical outcomes of HNS therapy in a Japanese cohort. METHODS:This study included 18 patients with moderate-to-severe OSA who underwent HNS implantation (Inspire UAS system®) at the study hospital between August 2022 and December 2024. Operative time, intraoperative blood loss, and perioperative complications were recorded to assess surgical efficiency. Sleep parameters, including apnea-hypopnea index (AHI), 3 % oxygen desaturation index (ODI), minimum SpO₂, time with SpO₂ < 90 %, Epworth Sleepiness Scale (ESS), Pittsburgh Sleep Quality Index (PSQI), and patient-reported impressions, were evaluated at 6 and 12 months post-implantation. RESULTS:The average operative time was 118.4 ± 48.6 min. Analysis of the surgical learning curve showed a significant decrease in operative time with increasing case number (ρ = -0.876, p < 0.01). Nonlinear regression using a logarithmic model indicated that operative time tended to decrease progressively with increasing surgical experience, with greater reductions observed in the early cases and more gradual decreases in later cases. Adverse events were infrequent, with one patient requiring device explantation. At 6 months, the mean AHI decreased from 30.8 events/hour (/h) to 9.5/h (p < 0.001), ODI from 25.9 /h to 7.2 /h (p < 0.001), ESS from 11.3 ± 6.0-7.3 ± 4.0 (p = 0.006), and PSQI from 11.4 ± 3.6-7.3 ± 3.1 (p = 0.008). These improvements were sustained at 12 months, with high patient satisfaction reported in ≥ 90 % of cases. CONCLUSION:HNS implantation is safe, and operative time decreases progressively with increasing experience. HNS therapy is effective and feasible for Japanese patients with OSA and produces substantial improvements in both objective sleep parameters and patient-reported outcomes. Despite the small sample size and short follow-up period, this pilot study provides important initial evidence supporting the applicability of HNS in Asian populations.

Behavioral hyperventilation–induced central sleep apnea (BH-CSA) is a recently recognized subtype of central sleep apnea (CSA). Although several cases have been reported recently, its long-term clinical course and reversibility remain unclear. Here, we report a case of BH-CSA that demonstrated reversibility over the long-term course. We report a previously healthy woman in her 50s with moderate obstructive sleep apnea (OSA; apnea–hypopnea index [AHI] 25.6 events/h) who developed severe BH-CSA during the COVID-19 pandemic. Pandemic-related psychological stress provoked repeated arousals with hyperventilation, during which CO₂ levels fluctuated and arterial PaCO₂ fell to a nadir of 31.7 mmHg, converting OSA to central sleep apnea (AHI 95 events/h, 93

Machine learning has been widely applied to sleep and arousal analysis using PSG (polysomnography), yet limited work has focused on in-home EEG (electroencephalography). Compared to sleep stage classification, arousal detection remains a more challenging task, often requiring specialized methods. Multitask learning offers a promising alternative by enabling models to learn from related tasks simultaneously, potentially improving arousal scoring performance. In this study, we explore multitask learning using different U-Net architectures for joint sleep and arousal scoring in in-home EEG data. We propose and evaluate two key architectural innovations: (1) the integration of bidirectional LSTM layers at different positions within the U-Net architecture, and (2) the use of a branching double decoder for task-specific outputs. The best-performing configurations for arousal scoring were SAAS U-Net DD-L and D-L (Double/Single Decoder followed by LSTM), while SAAS U-Net L-DD and L-D (LSTM between encoder and Double/Single Decoder) achieved superior results for sleep scoring. Notably, sleep scoring performance was further improved through fine-tuning strategies. Experiments across four datasets—including subjects with sleep apnea and low-noise EEG recordings—demonstrate the robustness and practical potential of the proposed models. These findings highlight the value of multitask learning and architectural enhancements to U-Net for advancing sleep diagnostics using portable EEG data.Clinical Relevance— This study highlights the potential of multitask learning models, enhanced with LSTM layers and double decoders, to improve the accuracy of sleep and arousal scoring using portable EEG data. These advancements could facilitate more reliable in-home sleep diagnostics, offering clinicians a practical and accessible tool for evaluating sleep disorders.

Sleep is influenced by environmental factors, and hot ambient temperature undermines sleep quality. Considering the global warming, it becomes increasingly important to keep bedroom temperature cool in the summer. Today, two electrical cooling systems are available: convection and radiant. Convection air cooling system supplies cooled air-flow. Radiant cooling system cools the room through radiation from the cooled surfaces, moisture condensation on which is prevented by reheated refrigeration cycle system. The present study was a randomized, cross-over trial setting room air temperature at 26 ℃ to compare subjective and objective quality of sleep between convection and radiant cooling in 7 healthy women. Subjective and objective quality of sleep were assessed by Oguri-Shirakawa-Azumi Sleep Inventory and EEG-based sleep parameters, respectively. Compared with convection cooling, radiant cooling provided lower relative humidity and air flow. Refreshness, one of the 4 indices of subjective quality of sleep, was higher with radiant cooling. Among the EEG-based sleep parameters, sleep efficiency was higher and sleep latency was shorter with radiant cooling. In spite of similar ambient room temperature, difference in cooling (radiant vs. convection) and/or dehumidifying (ordinary air conditioning vs. reheated refrigeration cycle) method provided difference in thermal comfort affecting subjective and objective quality of sleep.

Excessive daytime sleepiness (EDS) can impair athletic performance. In female athletes, sleep architecture changes in association with the menstrual cycle; however, studies examining the relationship between menstrual cycle and sleep, particularly EDS, remain limited. Therefore, we conducted subgroup analyses of our previous study, comparing changes in sleep measures associated with menses between athletes with and without EDS. Female collegiate athletes with regular menstrual cycles were recruited for this study. Participants underwent home electroencephalogram monitoring during the first and second nights after the onset of menses, and one night between the seventh and 10th nights after menses onset (mid-follicular phase). The Epworth Sleepiness Scale (ESS) was used to assess EDS. Interactions between the presence of EDS (i.e., ESS ≥ 11) and changes in objective measures of sleep in association with menses were analyzed. Data from 45 athletes, including 24 with EDS, revealed distinct changes in wake after sleep onset (WASO) among athletes with EDS compared with those without EDS (interaction p = 0.010). Specifically, athletes with EDS experienced increased WASO on the first night after the onset of menses compared with other nights (ANOVA p = 0.030). Pairwise comparisons showed significant differences between the first and second nights after menses onset (p = 0.035). Female collegiate athletes with regular menstrual cycles are more likely to experience EDS in association with increased WASO during the first night after the onset of menses.

Daytime napping improves performance, which is maximized with a post-N2 9-min nap. We evaluated whether a system that enables optimal-time automatic awakening using blood flow parameters could improve performance, sleepiness, and fatigue compared to no-nap. Additionally, we investigated whether its performance was comparable to manual awakening based on polysomnography. Eighty-one healthy adults (33.6 ± 12.8 years) were randomly assigned to automatic- or manual-awakening or rest groups. A task bout comprising a digit-symbol substitution test (DSST), visual detection test, and sleepiness and fatigue questionnaires was performed three times per session before napping and for six sessions after napping. In all post-nap sessions, sleepiness and fatigue in the automatic awakening group decreased, compared to the rest group, and were comparable to those in the manual awakening group. The DSST improved in the sixth post-nap session for the manual awakening group compared to the rest group; no improvement was observed in the automatic awakening group. The system model was refined by adding training data and tested on 50 healthy adults (40.6 ± 13.1 years). The test results revealed that the N2 detection accuracy of the system improved. The optimal automatic awakening system improves subjective sleepiness and fatigue, and further improvements in its accuracy may enhance post-nap performance.

Purpose:Home sleep apnea tests (HSATs) using polygraphy devices are becoming increasingly important for evaluating obstructive sleep apnea. Alice NightOne, a widely used polygraphy device, includes automatic scoring software; however, more reliable scoring results can be provided by incorporating advanced algorithmic systems like Somnolyzer. Despite this, the accuracy of automatic scoring of this polygraphy device using such applications has not been specifically investigated. Thus, in this study, we aimed to compare the respiratory event indices (REIs) obtained via automatic scoring versus manual scoring. Patients and Methods:Data of eligible patients who underwent HSAT with this polygraphy device were retrospectively analyzed using the following three methods: 1) manual scoring; 2) default automatic scoring of the analysis software; and 3) automatic scoring with the Somnolyzer system. The REIs were calculated using these three methods and expressed as mREI, aREI, and sREI, respectively. Correlations and agreements between the aREI, sREI, and mREI were assessed. Results:Data from 20 patients were analyzed. The mean mREI, aREI, and sREI were 14.7±13.3, 13.7±11.8, and 14.3±13.4 events/h, respectively. A strong correlation was found between aREI and mREI (coefficient, 0.976; P<0.01), with a mean difference between them of 1.0 and a limit of agreement of -5.3 to 7.3. The correlation between sREI and mREI was more prominent (coefficient, 0.996; P<0.001); their mean difference was 0.1, with a limit of agreement of -2.1 to 2.9. Conclusion:Automatic scoring of REI using this polygraphy device showed good correlation and agreement with manual scoring. The favorable correlation and agreement were more pronounced with the Somnolyzer system.

Automatic sleep stage classification is an important task to assist experts to perform diagnosis of sleep-related disorders. In supervised learning setting, the availability of a large amount of labeled training data is often the bottleneck for training the sleep stage classification model. Furthermore, the advent of multi-channel EEG signal from in-home EEG devices for sleep monitoring also pose a challenge of designing a label-efficient model suitable for such data. This work proposes a label-efficient approach leveraging self-supervised learning for performing 5 sleep stage classification suitable for multichannel in-home EEG device signal. Our work demonstrates the effectiveness of employing contrastive learning technique on unlabeled EEG data to learn the prominent features. We fine-tuned the features learned by contrastive learning to perform sleep stage classification using a limited amount of labeled data. The experimental results suggest that the proposed approach is more effective than training a traditional end-to-end sleep stage classification model without contrastively learned features. Specially, our approach is suitable in the situation where a large amount of unlabeled data is available and a small amount of labeled data is provided. Furthermore, our results suggest that the proposed method predicts with higher confidence than the end-to-end model.

Sleep insufficiency and sleep disorders pose serious health challenges. This study aimed to determine the potential discrepancy between subjective and objective sleep assessments, including the latter made by physicians, by analyzing a 421-participant dataset in Japan comprising multiple nights of in-home sleep electroencephalogram (EEG) data and questionnaire responses on sleep habits or subjective experiences. We employed logistic regression models to examine which subjective and objective sleep parameters physicians are paying attention to when assessing sleep insufficiency, insomnia, sleep quality, and sleep apnea. Questionnaire responses, including subjective sleep assessments, exhibited poor performance predicting physicians’ assessments, whereas objective data demonstrated good predictive performance, indicating a discrepancy between subjective and objective sleep assessments. Although the in-home sleep EEG measurements had minimal first night effects, incorporating measurements over multiple nights can improve the detection of objective insomnia. Moreover, we found that participants with severe sleep insufficiency overestimated their sleep duration, whereas those with subjective insomnia but without objective insomnia underestimated it. Additionally, subjective sleep quality reflected sleep efficiency but not the frequency of short awakenings or objective sleep depth. In particular, the effects of apnea on objective sleep quality were not subjectively perceived. Collectively, our findings suggest that subjective sleep assessments alone are insufficient for evaluating sleep health and that health checkups and advice based on sleep EEG measurements may be useful in improving sleep habits and for early detection of sleep disorders.

With the rising awareness of the critical role sleep plays in both health and social well-being, the demand for sleep studies is rapidly increasing.Automatic sleep stage classification is a fundamental part of sleep measurement, and machine learning models have been developed to assist in this process. These models achieve accuracy comparable to that of technicians when using data from healthy individuals. However, sleep patterns in individuals with sleep disorders, such as sleep apnea syndrome (SAS), one of the most common sleep disorders, differ from those of healthy individuals. As a result, existing models trained on healthy individuals’ data do not achieve sufficient accuracy when applied to SAS patients. This is a barrier to clinical application.A recent study using in-home EEG devices showed that technicians can accurately classify sleep stages in SAS cases by considering surrounding epochs. Based on this, we developed a model dedicated to SAS patients that incorporates the temporal context of relevant epochs.We found that this context-aware model significantly improved classification accuracy compared to models that only focused on the target epoch. In the training process using data from 76 severe SAS cases, the model based solely on single-epoch data achieved an accuracy of 71.5%, while the model considering the surrounding epochs achieved an accuracy of 73.7%. The classification accuracy improved across all stages except N3.This approach appears to capture the frequent sleep stage transitions characteristic of SAS.

Efforts to simplify standard polysomnography (PSG) in laboratories, especially for obstructive sleep apnea (OSA), and assess its agreement with portable electroencephalogram (EEG) devices are limited. We aimed to evaluate the agreement between a portable EEG device and type I PSG in patients with OSA and examine the EEG-based arousal index’s ability to estimate apnea severity. We enrolled 77 Japanese patients with OSA who underwent simultaneous type I PSG and portable EEG monitoring. Combining pulse rate, oxygen saturation (SpO2), and EEG improved sleep staging accuracy. Bland–Altman plots, paired t-tests, and receiver operating characteristics curves were used to assess agreement and screening accuracy. Significant small biases were observed for total sleep time, sleep latency, awakening after falling asleep, sleep efficiency, N1, N2, and N3 rates, arousal index, and apnea indexes. All variables showed > 95% agreement in the Bland–Altman analysis, with interclass correlation coefficients of 0.761–0.982, indicating high inter-instrument validity. The EEG-based arousal index demonstrated sufficient power for screening AHI ≥ 15 and ≥ 30 and yielded promising results in predicting apnea severity. Portable EEG device showed strong agreement with type I PSG in patients with OSA. These suggest that patients with OSA may assess their condition at home.

Manual sleep and arousal scoring is a labor-intensive task that demands significant time and effort. To speed up this process, several automatic scoring models based on deep learning have been proposed. These models primarily focus on scoring PSG (Polysomnogram) signals by separately classifying sleep stages and arousal events. This study introduces a novel methodology for concurrent sleep stage classification and arousal scoring, employing multitask learning for the analysis of in-home EEG (Electroencephalogram) signals. Our approach led to improvements in overall precision and sensitivity of arousal scoring, with values increasing by 0.3% to 4%. Notably, this approach did not yield improvements in sleep scoring. We validated our methodology on two private datasets collected from in-home IoT (internet of Things) EEG devices and achieved consistent outcomes. Collectively, our research underscores the benefits of multitask learning for arousal scoring in in-home EEG signals.

Purpose:Female athletes with menstrual abnormalities have poor sleep quality. However, whether female athletes with poor sleep quality based on subjective assessment have distinctive changes in objective measures of sleep in association with menses remains unclear. This study aimed to compare changes in objective sleep measurements during and following menses between collegiate female athletes with and without poor subjective sleep quality.Patients and Methods:Female collegiate athletes (age range/mean ± standard deviation: 18-22/ 22.2±1.1) with regular menstrual cycles were recruited. The participants underwent home electroencephalogram monitoring during the first and second nights after the onset of menses and one night between the seventh and 10th nights after menses onset (mid-follicular phase). The Pittsburgh Sleep Quality Index (PSQI) was used to assess the subjective sleep quality. Interactions between the presence of poor subjective sleep quality (ie, PSQI ≥6) and changes in objective measures of sleep in association with menses were analyzed.Results:Data of 45 athletes, including 13 with poor subjective sleep quality, showed that changes in arousal index in athletes with poor subjective sleep quality were distinctive from those in athletes without poor subjective sleep quality (p = 0.036 for interaction). In athletes with poor subjective sleep quality, the arousal index was significantly increased in menses (p for analysis of variance, 0.015), especially on the first night after the onset of menses compared with during the mid-follicular phase (p = 0.016).Conclusion:Collegiate female athletes with regular menstrual cycles are likely to have poor subjective sleep quality in association with more frequent arousal during the first night after the onset of menses than during the mid-follicular phase.

In-home automated scoring systems are in high demand; however, the current systems are not widely adopted in clinical settings. Problems with electrode contact and restriction on measurable signals often result in unstable and inaccurate scoring for clinical use. To address these issues, we propose a method based on ensemble of small sleep stage scoring models with different input signal sets. By excluding models that employ problematic signals from the voting process, our method can mitigate the effects of electrode contact failure. Comparative experiments demonstrated that our method could reduce the impact of contact problems and improve scoring accuracy for epochs with problematic signals by 8.3 points, while also decreasing the deterioration in scoring accuracy from 7.9 to 0.3 points compared to typical methods. Additionally, we confirmed that assigning different input sets to small models did not diminish the advantages of the ensemble but instead increased its efficacy. The proposed model can improve overall scoring accuracy and minimize the effect of problematic signals simultaneously, making in-home sleep stage scoring systems more suitable for clinical practice.

Suvorexant is an orexin receptor antagonist that targets the wake-promoting system. Orexin is also known to regulate energy metabolism in rodents, but its role in humans remains largely unknown. Here, we assessed the effect of suvorexant (20 mg) on energy metabolism during sleep and shortly after awakening in a randomized, double-blind, placebo-controlled, crossover study in 14 healthy men. Suvorexant increased rapid eye movement (REM) but decreased nonrapid eye movement (NREM) stage 1. Energy expenditure during wake after sleep onset (WASO) was higher than that during NREM and REM sleep in the placebo but not in the suvorexant trial, suggesting that the increase in energy expenditure during WASO was due to an activation of the orexin system. Fat oxidation during sleep increased, and its effect remained after waking the next morning. Suvorexant decreased protein catabolism but did not affect overall energy expenditure. The orexin system may affect fat oxidation independent of its roles in sleep regulation in humans.