Organizations: Hong Kong Xsmart Century Technology Co., Ltd., Hong Kong, China · Urban Mobility Lab, Massachusetts Institute of Technology, Cambridge, MA 02139, USA · Singapore-MIT Alliance for Research and Technology (SMART) Centre, Singapore 138602, Singapore · Department of Respiratory and Critical Care Medicine, Binzhou Medical University Hospital, Binzhou 256603, China · Interdisciplinary Center of Sleep Medicine, Charit´e – Universitätsmedizin Berlin, 10117 Berlin, Germany
Sleep staging is essential for the assessment of sleep quality and the diagnosis of sleep-related disorders. Conventional polysomnography (PSG), while considered the gold standard, is intrusive, labor-intensive, and unsuitable for long-term monitoring. This study evaluates the performance of the Sleepal AI Lamp, a contactless, radar-based consumer-grade sleep tracker, in comparison with gold-standard polysomnography (PSG), using a large-scale dataset comprising 1022 overnight recordings. We extract multi-scale respiratory and motion-related features from radar signals to train a frequency-augmented deep learning model. For the binary sleep-wake classification task, experimental results demonstrated that the model achieved an accuracy of 92.8% alongside a macro-averaged F1 score of 0.895. For four-stage classification (wake, light NREM (N1 + N2), deep NREM (N3), REM), the model achieved an accuracy of 78.5% with a Cohen's kappa coefficient of 0.695 in healthy individuals and maintained a stable accuracy of 77.2% with a kappa of 0.677 in a heterogeneous population including patients with varying severities of obstructive sleep apnea (OSA). These experimental results demonstrate that the sleep staging performance of the contactless Sleepal AI Lamp is in high agreement with expert-labeled PSG sleep stages. Our findings suggest that non-contact radar sensing, combined with advanced temporal modeling, can provide reliable sleep staging performance without requiring physical contact or wearable devices. Owing to its unobtrusive nature, ease of deployment, and robustness to long-term use, the contactless Sleepal AI Lamp shows strong potential for clinical screening, home-based sleep assessment, and continuous longitudinal sleep monitoring in real-world medical and healthcare applications.
Figures & tables
Figure 1: Physical prototype of the Sleepal AI Lamp. The transparent black section at the top houses the core data acquisition module.
Characteristics
Value (Mean ± S.D. or Count)
Gender (Male/Female)
670 / 352
Age (years)
44.01 ± 15.88
BMI (kg/m 2 )
26.80 ± 5.43
AHI (events/h)
14.01 ± 21.76
Time in Bed (TIB, min)
534.70 ± 61.82
Total Sleep Time (TST, min)
409.30 ± 61.98
Table 1: Demographic Characteristics and Sleep Data of Participants.
Figure 2: Time-synchronized visualization of radar-derived physiological signals and the corresponding expert-annotated sleep hypnogram over a full night. From top to bottom, the panels display the continuous respiratory rate variation, body movement amplitude, and the reference sleep stages. Dashed red vertical lines highlight representative segments of Wake, Deep, REM, and Light sleep, illustrating the distinct respiratory and motor patterns characteristic of each stage.
Figure 3: Visualization of the chronobiological and temporal feature embeddings. (a) Circadian phase encoding using a phase-shifted cosine function. (b) Normalized linear temporal feature representing homeostatic sleep progression. (c) Typical probability distribution of sleep stages across the night.
Figure 4: Schematic overview of the proposed sleep staging architecture.
Figure 5: Letter-value plot visualizing the session-level performance for binary sleep-wake classification across the validation cohort. Sleep sensitivity (left) and sleep specificity (right) are computed for each subject. In this visualization, the center line denotes the median, and the primary boxes immediately above and below represent the interquartile range (IQR). Successive boxes partition the remaining percentile ranges in half at each step, providing a detailed view of the distribution tails, with outliers represented as individual points.
Figure 6: Evaluation of sleep onset prediction. (A) Scatter plot of per-subject time difference (Prediction - True) with the ±1 standard deviation (SD) band shaded. (B) Histogram and density curve of the error distribution.
Figure 7: Evaluation of sleep offset prediction. (A) Scatter plot of per-subject time difference (Prediction - True) with the ±1 standard deviation (SD) band shaded. (B) Histogram and density curve of the error distribution.
Figure 8: Confusion matrix of the four-class sleep staging across the independent validation set.Each matrix is row-normalized to represent the recall (sensitivity) of each sleep stage, with values indicating the percentage of 30-s epochs.
Figure 9: Subject-level performance distributions for Accuracy, F1 Score, and Cohen’s Kappa across the validation cohort ( N=260 ). Left panels show histograms with kernel density estimate (KDE) curves, while right panels present horizontal letter-value plots (boxenplots) visualizing distribution density and tails.
Figure 10: Bland-Altman plots assessing the agreement between predicted and ground-truth sleep stage durations (Wake, Light, Deep, and REM).
Stage
True (mean ± std)
Pred (mean ± std)
Diff (mean ± std)
Wake
125.4 ± 66.7
120.4 ± 48.6
−5.0±30.6
Light
225.4 ± 60.8
220.2 ± 50.7
−5.2±44.7
Deep
89.1 ± 44.4
90.6 ± 29.1
+1.5±44.1
REM
94.8 ± 34.7
103.6 ± 29.9
+8.8±24.5
Table 2: Comparison of sleep stage durations (minutes) between label and model predictions.
Figure 11: Comparison of mean sleep stage durations between PSG annotations (Label) and model predictions.
AHI Group
N
Accuracy (%)
F1 (macro)
Cohen’s Kappa
Normal (AHI ≤ 5)
124
78.5 ± 6.1
0.777 ± 0.079
0.695 ± 0.091
Mild ( 5<AHI≤15 )
55
75.3 ± 5.4
0.743 ± 0.057
0.648 ± 0.072
Moderate ( 15<AHI≤30 )
38
77.9 ± 5.9
0.767 ± 0.063
0.676 ± 0.087
Severe ( AHI>30 )
43
74.3 ± 7.6
0.711 ± 0.122
0.614 ± 0.137
Table 3: Subject-level classification performance across different AHI severity groups (mean ± std).
Figure 12: Normalized confusion matrices for the four-stage sleep classification stratified by Obstructive Sleep Apnea (OSA) severity. The four groups are defined by the Apnea-Hypopnea Index (AHI): Normal, Mild, Moderate, and Severe. Each matrix is row-normalized to represent the recall (sensitivity) of each sleep stage, with values indicating the percentage of 30-s epochs.
Metric
SE ≥ 80%
60% ≤ SE < 80%
SE < 60%
( N=156 )
( N=82 )
( N=22 )
Accuracy (%)
76.7 ± 5.8
76.4 ± 6.2
76.3 ± 5.2
Cohen’s Kappa
0.668 ± 0.084
0.655 ± 0.095
0.651 ± 0.079
Macro-averaged F1
0.751 ± 0.071
0.754 ± 0.076
0.723 ± 0.069
F1 Wake
0.841 ± 0.088
0.786 ± 0.130
0.862 ± 0.065
F1 Light
0.725 ± 0.098
0.749 ± 0.079
0.658 ± 0.120
Table 4: Sleep staging performance across three subgroups stratified by Sleep Efficiency (SE) (mean ± std).
Automated sleep staging is commonly approached as a supervised machine learning problem, with deep learning methods dominating recent research. While machine learning models achieve near-human level agreement with human-scored reference sleep stages, their decisions are typically opaque and not designed to follow clinical scoring rules. We propose a transparent alternative: a deterministic, rule-based sleep staging method that explicitly operationalizes the American Academy of Sleep Medicine's (AASM) scoring logic as executable code, coupled with epoch-level natural-language justifications derived from an explanation trace. We evaluate the approach on 50 polysomnography recordings with a 10-scorer majority-vote consensus as reference. Across all recordings, the method agreed with the majority-vote reference in 60.5% of epochs (κ=0.42), with substantially higher agreement on a dataset used during development (77.1%, κ=0.61). Agreement with the reference was highest for sleep stage N2 (recall 83.5%) and moderate for sleep stage R (recall 68.7%), while Wake and N1 recall were low. Despite lower agreement with the reference than contemporary deep learning models, the method provides deterministic decisions and natural language explanations aligned with AASM scoring rules, making it a complementary tool for auditing, debugging, and governing deep learning-based sleep staging.
Emil Hardarson, Konstantin Popov, Sigridur Sigurdardottir +3
Department of Computer Science, Reykjavik University, Reykjavik, Iceland · Reykjavik University Sleep Institute, Reykjavik University, Reykjavik, Iceland · Department of Engineering, Reykjavik University, Reykjavik, Iceland +1
Automatic sleep stage scoring is crucial for the diagnosis and treatment of sleep disorders. Although deep learning models have advanced the field, many existing models are computationally demanding and designed for single-channel electroencephalography (EEG), limiting their practicality for multimodal polysomnography (PSG) data. To overcome this, we propose ULW-SleepNet, an ultra-lightweight multimodal sleep stage scoring framework that efficiently integrates information from multiple physiological signals. ULW-SleepNet incorporates a novel Dual-Stream Separable Convolution (DSSC) Block, depthwise separable convolutions, channel-wise parameter sharing, and global average pooling to reduce computational overhead while maintaining competitive accuracy. Evaluated on the Sleep-EDF-20 and Sleep-EDF-78 datasets, ULW-SleepNet achieves accuracies of 86.9% and 81.4%, respectively, with only 13.3K parameters and 7.89M FLOPs. Compared to state-of-the-art methods, our model reduces parameters by up to 98.6% with only marginal performance loss, demonstrating its strong potential for real-time sleep monitoring on wearable and IoT devices. The source code for this study is publicly available at https://github.com/wzw999/ULW-SLEEPNET.
Zhaowen Wang, Dongdong Zhou, Qi Xu +3
School of Computer Science and Technology, Dalian University of Technology, Dalian, China · Faculty of Information Technology, University of Jyväskylä, Jyväskylä, Finland · Key Laboratory of Social Computing and Cognitive Intelligence, Ministry of Education, Dalian, China +3
Sleep is essential for health, yet studying its dynamics requires manual sleep staging, a labor-intensive step in research and clinical care. Across centers, polysomnography (PSG) recordings are traditionally scored in 30-s epochs for pragmatic, not physiological, reasons and vary in electrode count, montage, and subject characteristics. These constraints challenge harmonized multi-center studies and the discovery of robust biomarkers on shorter timescales. We present AnySleep, a deep neural network that scores sleep from any electroencephalography (EEG) or electrooculography (EOG) data at adjustable temporal resolutions. We trained and validated the model on over 20,000 overnight recordings (> 200,000 hours of EEG and EOG) from 28 datasets across multiple clinics to promote robust generalization across sites. The model attains state-of-the-art performance and surpasses or equals established baselines at 30-s epochs. Performance improves with more channels, yet remains strong when EOG is absent or only EOG or single EEG derivations (frontal, central, or occipital) are available. On sub-30-s timescales, the model captures short wake intrusions consistent with arousals and improves prediction of pathophysiological conditions (obstructive sleep apnea, narcolepsy type 1, insomnia) over 30-s scoring. We make the model publicly available to facilitate large-scale studies with heterogeneous electrode setups and accelerate biomarker discovery in sleep.
Department of Medical Engineering and Technomathematics, FH Aachen University of Applied Sciences, 52428 Jülich, Germany · Department of Information and Computing Sciences, Utrecht University, Utrecht, The Netherlands · Institute for Data-Driven Technologies, FH Aachen University of Applied Sciences, 52428 Jülich, Germany