Pith. sign in

REVIEW 3 major objections 4 minor 46 references

Rethinking PPG-based Sleep Staging: Datasets, Metrics, and Benchmarks

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read PPG sleep staging lags EEG because the 30-second epoch label averages away the signal's sharpest, boundary-concentrated information; second-level supervision closes part of that gap.

desk verdict A useful but overclaimed paper: the 3.7–5.7 pp gain is real against independent labels, but the missing epoch-collapsed pseudo-label ablation conflates sec-level supervision with label denoising. read the letter →

arxiv 2608.00943 v1 pith:JKR5YGEY submitted 2026-08-02 cs.HC cs.LG

Shuntian Zheng , Jiawei Wang , Cong Fu , Huan Yu , Chen Chen , Yu Guan , Sai Gu This is my paper
classification cs.HCcs.LG
keywords sleepstagingphotoplethysmographyboundarydetectionlabelexpansionhiddensemi-Markovmodelwearablemonitoringtransitionzero-shottransfer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that the performance gap between PPG-based and EEG-based sleep staging is mostly a mismatch between signal and task, not a lack of information in PPG. PPG's cardiovascular features—heart-rate variability and pulse morphology—change sharply within seconds at sleep-stage boundaries, while inside a stable stage they differ between stages only subtly; a 30-second epoch label averages over exactly the informative boundary region. The authors build a label-expansion pipeline that converts coarse 30-second PSG epoch labels into per-second pseudo-labels, validate those labels against physician-reviewed sec-level annotations, and show that using them as supervision improves conventional four-class epoch-level staging by 3.7–5.7 percentage points in accuracy across six evaluated architectures on MESA, with the benefit persisting in zero-shot transfer to CFS. The proposed framing is that sleep staging for PPG should be treated as a transition-detection problem with boundary-aware supervision and boundary-aware metrics, rather than as epoch classification alone.

What carries the argument

The central object is the label-expansion pipeline that produces sec-level SSTD supervision from coarse 30-second epoch labels. It combines a Hidden Semi-Markov Model with explicit stage-duration distributions and soft constraints from the coarse labels, followed by local changepoint refinement that repositions each detected boundary to maximize evidence of a distributional shift. A physiological post-processing step enforces that direct Wake–Deep transitions do not occur by inserting brief Light-sleep segments. The pipeline's output is used both as training supervision and as the reference for two new metrics, Boundary Error (BE) and Transition F1 (TF1), which measure temporal localization

What would settle it

Have sleep physicians annotate second-level transition onsets on a random subset of MESA or CFS recordings and compare the pipeline's boundaries against those target-cohort expert labels. If the boundary error is substantially larger than the 4.2 s reported on the separate validation set, or if models trained on the expert labels fail to reproduce the 3.7–5.7 pp epoch-accuracy gains, the central transfer claim is refuted.

Watch

Extended reading notes

Core claim

The central discovery is a characteristic asymmetry: under HRV and pulse-morphology features, PPG has higher boundary-to-stable contrast than EEG (normalized variance ratio 1.84 vs 1.31) but lower inter-stage separability inside stable epochs (η² = 0.07 vs 0.44). The paper's claim is that PPG's useful sleep information is concentrated at stage transitions, so the conventional 30-second epoch classification paradigm suppresses it. To act on this, the paper defines Sleep Stage Transition Detection (SSTD): predicting the onset time, from-stage, and to-stage of each transition at second resolution. It then constructs sec-level pseudo-labels for public PSG cohorts with a two-stage HSMM-plus-chang

Load-bearing premise

The load-bearing premise is that the per-second pseudo-labels generated by the expansion pipeline are accurate enough on MESA and CFS to serve as supervision, even though the only direct check against expert second-level annotations was done on a separate 50-recording dataset; accuracy on the target cohorts is inferred, not measured.

Editorial extensions

If this is right

  • Sec-level supervision transfers to conventional epoch staging: every matched architecture improves after SSTD pre-training, and the gains are measured against the original PSG epoch labels, which are independent of the pseudo-label pipeline.
  • Boundary-aware metrics expose differences that epoch accuracy hides: SleepPPGNet and PPG-TCN have similar epoch accuracy (74.1 vs 73.6) but differ in transition F1 (46.5 vs 49.1) and boundary error (7.4 vs 6.8 s).
  • The supervision benefit does not require a strong model: LightSleep, a deliberately simple 1D CNN, achieves the best transition localization (BE 4.6 s, TF1 59.4%) and improves epoch accuracy from 72.4 to 77.2.
  • The framework degrades under the full five-class AASM taxonomy (TF1 drops 11.1 pp relative to four-class), so the four-class vocabulary is the empirically reliable setting for PPG-only SSTD.
  • The label information transfers to a different task: SSTD pre-training improves sleep-wake detection accuracy by about 4–5 pp over epoch-only training on the same architectures.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's 4.2 s mean boundary error on the expert validation set is close to the 3.8 s mean inter-rater disagreement among physicians; one consequence, not drawn by the paper, is that further label refinement may hit the ceiling of expert agreement, so the next bottleneck could be defining a consensus reference rather than improving the pipeline.
  • Because the pipeline is described as applicable to any PSG-scored dataset with concurrent PPG and requires no dataset-specific architecture, the same boundary-aware supervision could be applied to ECG or other coarse-labeled physiological signals; the paper notes ECG adaptation as future work but does not test it.
  • A direct test of the mechanism would be to jitter the boundary positions in the pseudo-labels during training; if epoch-level gains persist under random jitter, the benefit would come from finer granularity rather than accurate boundary timing, which the paper does not run.
  • The physiological constraint that direct Wake–Deep transitions never occur is imported from normal adult sleep; in populations with disrupted architecture (narcolepsy, severe apnea, pediatric sleep) the constraint may insert false Light segments, so the reported gains may not transfer to those groups without relaxing it.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The paper argues that PPG-based sleep staging underperforms EEG-based staging because the conventional 30-second epoch classification framework discards the sec-level boundary information in which PPG is most informative. It proposes Sleep Stage Transition Detection (SSTD), a transition-localization task, and a two-stage label-expansion pipeline (HSMM plus local changepoint refinement with physiological constraints) that converts coarse PSG epoch labels into sec-level pseudo-labels under a four-class vocabulary (Wake, Light, Deep, REM). The pipeline is validated on a separate expert-reviewed dataset (kappa=0.81, TF1=78.6%, BE=4.2s), on MESA and CFS via window consistency, and through a sleep-wake transfer task. The central empirical claim is that training on these sec-level pseudo-labels and collapsing predictions to 30-second epochs improves four-class epoch-level staging against the original PSG epoch labels on MESA by 3.7--5.7 pp in accuracy across six baselines, with zero-shot transfer to CFS. The epoch-level evaluation uses original PSG labels, independent of the pseudo-label pipeline, which is a genuine strength. However, the causal attribution of the gain to sec-level granularity is not fully isolated, because the pipeline also changes the label source at epoch level, and the reported improvements are point estimates from a single seed without uncertainty quantification.

Significance. If the transfer result holds, the paper offers a practical and label-efficient route to improving PPG sleep staging without new expert sec-level annotations, and it introduces a useful task formulation and metrics (BE, TF1) that expose information invisible to epoch-level accuracy. Credit is due for several strengths: the epoch-level transfer gains are measured against original PSG labels independent of the expansion pipeline; the gains are consistent across six architectures and in zero-shot CFS evaluation; the pipeline is externally checked by physician review; and an auxiliary sleep-wake task provides a non-circular functional validation. These elements make the central claim credible. The main weaknesses are the missing ablation that separates label-source changes from granularity changes, and the absence of confidence intervals or significance tests for the headline accuracy differences. Both are addressable and do not invalidate the approach, but they are load-bearing for the abstract's causal claim.

major comments (3)
  1. The abstract attributes the epoch-level accuracy gains to sec-level supervision, but the comparison in Table 6 simultaneously changes two factors: label source (original 30-s PSG labels vs. the pipeline's pseudo-labels) and temporal granularity (epoch vs. second). Table 2 shows that the pseudo-labels have only 0.857 window consistency with the original MESA labels, so training on them is also training on a partially relabeled/denoised set. Table 8 compares full-pipeline labels with and without changepoint refinement, which holds granularity fixed and varies label precision; it does not include the condition of training on the same pseudo-labels collapsed back to 30-s epochs by majority vote. Without that condition, the 3.7--5.7 pp gains could arise from label denoising alone rather than from boundary-timing information. Please add an ablation that trains on epoch-collapsed pseudo-labels
  2. All transfer numbers in Table 6 (and in Appendix E) are point estimates from a single random seed (seed 42) with no confidence intervals, standard deviations, or significance tests. The claimed improvements of 3.7--5.7 pp in accuracy and up to 6.2 pp in MF1 could be within run-to-run variability, especially given the modest absolute differences. Please report multi-seed means with standard deviations or bootstrap confidence intervals, and, where feasible, paired tests across subjects. This is a necessary support for the headline quantitative claim.
  3. The SSTD benchmark and the pseudo-label training on MESA/CFS are evaluated against labels produced by the same expansion pipeline. The external expert review covers only a separate 50-recording set; the only on-target checks are window consistency and implausible-transition rate against the same PSG labels the pipeline was designed to match. In addition, the physiological constraint inserts artificial Light segments at Wake--Deep boundaries, and the inserted dwell time is a free parameter whose effect on boundary localization is not analyzed on MESA/CFS. The paper acknowledges in Section H that SSTD metrics quantify agreement with pseudo-labels rather than gold sec-level annotations, but the practical risk remains. Please report sensitivity of the downstream transfer gains and transition metrics to the inserted Light-segment duration, and ideally obtain sec-level expert labels on a small
minor comments (4)
  1. The abstract states 'four architecturally diverse baselines,' but Table 6 reports six methods. The conclusion also reports MF1 gains of 1.7--4.7 pp while the abstract emphasizes accuracy gains of 3.7--5.7 pp; please reconcile these numbers and be explicit about which metric is the headline.
  2. The caption is garbled: 'LightSleep is trained against the expanded SSTD labels first, then trained and evaluated with other models against MESA's original 30s labels' conflicts with the surrounding text, which states that Table 5 is evaluated under the expanded-label setting. Please rephrase to describe which models are SSTD-trained and which are epoch-only.
  3. The sentence 'Window consistency measures 30-second coarse agreement and functions as a lower bound on label fidelity' is imprecise. Majority-vote agreement with the original coarse labels is not formally a lower bound on sec-level fidelity; please rephrase as a descriptive consistency statistic.
  4. Cohen's kappa is first used without definition. State that it is epoch-level agreement after majority-vote collapse, and specify whether the reported expert inter-rater kappa is also computed in the same four-class setting.

Circularity Check

2 steps flagged · score 4.0 of 10

Partial circularity in the label-quality validation and the SSTD benchmark reference; the central epoch-level transfer claim is externally anchored but the sec-level attribution is partly confounded.

  1. self definitional [Section 3.3 Eq. (1) and Section 4.2 'Label Expansion Validation']
    "coarse epoch-level sleep stage labels enter as multiplicative soft constraints: for each second t, the emission likelihood P(x_t|s_t) is scaled by λ when s_t agrees with the enclosing 30-second epoch label ℓ_t and by 1/λ otherwise, with λ=3 ... Table 2 reports two automatic quality metrics computed against the original PSG-scored epoch labels, which are independent of the expansion pipeline."

    The HSMM in Eq. (1) takes the original epoch labels ℓ_t as direct inputs (soft constraints). The 'window consistency' metric then measures the fraction of 30-second windows in which the pseudo-label majority agrees with those same ℓ_t. High agreement is therefore partly enforced by construction, and calling the original labels 'independent of the expansion pipeline' is inaccurate. This does not make the entire paper circular because external expert review and the sleep-wake task provide independent checks, but this specific validation claim is self-referential.

  2. self definitional [Section 3.2 'Evaluation metrics' and Section 4.3 'Transition-Level Analysis on MESA']
    "In the present benchmark, the sec-level reference events are produced by the label expansion pipeline described in Section 3.3 and externally checked against expert review before downstream use. ... Table 5 provides a transition-level analysis under the expanded-label setting."

    The SSTD metrics (BE, TF1) are evaluated against reference events generated by the same label expansion pipeline that produces the sec-level training targets. On MESA and CFS, no expert sec-level labels exist, so the benchmark reference is pipeline-internal; the expert review was performed only on a separate 50-recording validation set. Consequently, the SSTD leaderboard partly measures agreement with the pipeline's own outputs rather than with an independent sec-level ground truth. The epoch-level transfer comparison in Table 6 uses original PSG labels, so the central staging claim is not entirely circular, but the SSTD benchmark itself is partially self-referential.

full rationale

The paper's central practical claim—that sec-level supervision improves conventional epoch-level staging—is evaluated against the original PSG epoch labels (Table 6) and under zero-shot transfer to CFS (Appendix E), so it is not forced by construction. Independent external anchors exist: expert physician review on a separate dataset (κ=0.81, TF1=78.6%) and an auxiliary sleep-wake task whose labels are independent of the pipeline. However, two components of the validation chain are self-referential. First, the 'window consistency' metric in Table 2 is computed against the same coarse epoch labels that serve as soft constraints in Eq. (1); the paper's claim that these labels are 'independent of the expansion pipeline' is not correct, and the reported 0.857 partly reflects constraint enforcement. Second, the SSTD benchmark (BE/TF1 in Tables 5 and 11) uses as reference the pipeline-generated sec-level labels used for training, with expert review only on a separate 50-recording set; on the main cohorts the reference is pipeline-internal. These issues do not invalidate the epoch-level transfer result, but they mean the abstract's causal attribution to 'sec-level supervision' is not cleanly separated from label denoising. The missing control of training on the same pseudo-labels collapsed to 30-second epochs is a confound rather than a circular reduction, so the score is moderate rather than severe.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the fidelity of pseudo-labels produced by the HSMM plus changepoint pipeline. That fidelity inherits all of the modeling assumptions above: Poisson durations, the physiological transition constraint, the soft coarse-label weighting, the changepoint heuristic, and the four-class vocabulary. No new physical entities are introduced.

free parameters (6)
  • lambda (label-consistency weight) = 3
    Scales HSMM emission likelihood when the inferred sec-level state agrees with the coarse epoch label; sensitivity tested in Appendix D.1, chosen by hand.
  • Delta (changepoint search radius) = 15 s
    Local window for boundary refinement in Eq. 2; chosen by hand, sensitivity reported.
  • beta (variance-asymmetry weight) = 0.5
    Weights post-transition variance increase in Eq. 2; chosen by hand.
  • delta_match (matching tolerance for TF1/BE) = 10 s
    Tolerance for matching predicted to reference transitions; main text choice, sensitivity in Appendix D.3.
  • inserted Light-stage dwell time = minimum credible Light dwell time from training population
    Duration of Light segment inserted at Wake-Deep boundaries by the physiological constraint.
  • HSMM transition matrix and Poisson duration parameters = estimated from 30s label sequences
    Fitted to the observed coarse labels in each cohort; a modeling choice about the generative process.
assumptions (5)
  • domain assumption Sleep stage durations follow Poisson distributions with stage-dependent rates
    Invoked in Section 3.3 Stage 1 for the HSMM; if real dwell times are not Poisson, the pseudo-label boundaries inherit bias.
  • domain assumption Direct Wake-Deep and Deep-Wake transitions do not occur in normal adult sleep; inserting a Light segment restores physiological plausibility
    Section 3.3 Physiological transition constraints; this hard rule fabricates Light segments when the data contradict it, which could corrupt labels in sleep-disordered populations.
  • domain assumption The soft coarse-label constraint with lambda=3 does not dominate the likelihood, allowing sec-level labels to deviate near boundaries
    Eq. 1; if lambda is too strong, pseudo-labels simply replicate epoch labels; if too weak, they lose the anchor. The chosen value is a modeling judgment.
  • domain assumption The changepoint objective in Eq. 2, with 10s windows and beta=0.5, identifies the true transition onset within a 15s window
    Section 3.3 Stage 2; the objective is a heuristic for distributional shift and is not derived from physiological transition dynamics.
  • domain assumption The four-class vocabulary (Wake, Light=N1+N2, Deep=N3, REM) preserves clinically relevant structure for PPG
    Section 3.2; merging N1/N2 is a modeling choice justified by PPG limitations, and Appendix F shows degradation when separating them.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Rethinking PPG-based Sleep Staging: Datasets, Metrics, and Benchmarks." pith.science (2026). https://pith.science/paper/JKR5YGEY

@misc{pith2026260800943,
  author       = {Pith},
  title        = {Pith review of: Rethinking PPG-based Sleep Staging: Datasets, Metrics, and Benchmarks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JKR5YGEY}},
  note         = {Machine review of arXiv:2608.00943}
}
read the original abstract

Automated sleep staging assigns discrete stage labels to successive time epochs throughout an overnight recording; conventionally each window spans at least 30 seconds, reflecting the minimum temporal resolution of the clinical scoring standard. Wearable photoplethysmography (PPG) has attracted sustained interest as an ambulatory alternative to laboratory-based polysomnography, which relies on electroencephalography (EEG) and other recording modalities that are impractical outside clinical environments. Yet PPG-based staging trails EEG-based methods by a substantial margin, and we argue this gap largely reflects a mismatch between signal and task. Within a stable stage, PPG's inter-stage feature differences are more subtle than those in EEG; yet at stage boundaries, PPG's principal cardiovascular features, heart rate variability and pulse morphology, shift sharply within seconds. The conventional practice of assigning one label to each 30-second epoch therefore suppresses feature that is concentrated near boundaries. We address this gap in two steps. First, we develop a label expansion pipeline based on Hidden Semi-Markov Models that converts coarse epoch labels into sec-level annotations. To assess whether these expanded labels are reliable enough for downstream supervision, we validate them on a separate expert-reviewed dataset and through an auxiliary sleep-wake task whose labels are independent of the expansion pipeline. Second, we use the resulting sec-level supervision on MESA to improve conventional four-class epoch-level staging across four architecturally diverse baselines by 3.7--5.7\,pp in accuracy against the original epoch labels, with supplementary zero-shot evaluation on CFS showing that the transfer benefit persists under cohort and annotation-protocol shift.

Figures

Figures reproduced from arXiv: 2608.00943 by the authors.

Figure 1
Figure 1. Illustrative EEG and PPG signals from continuous sleep segments. EEG exhibits clearer [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the annotation gap addressed by this work. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Population-level feature distributions on separate MESA subjects for EEG (left) and [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

46 extracted references · 41 canonical work pages

  1. [1]

    Disorders of sleep: an overview.Primary care, 32(2):305, 2005

    Leon Ting and Atul Malhotra. Disorders of sleep: an overview.Primary care, 32(2):305, 2005

    work page 2005

  2. [2]

    International classification of sleep disorders.Chest, 146(5):1387–1394, 2014

    Michael J Sateia. International classification of sleep disorders.Chest, 146(5):1387–1394, 2014

    work page 2014

  3. [3]

    The aasm manual for the scoring of sleep and associated events

    Richard B Berry, Rita Brooks, Charlene E Gamaldo, Susan M Harding, Carole Marcus, Bradley V Vaughn, et al. The aasm manual for the scoring of sleep and associated events. Rules, Terminology and Technical Specifications, Darien, Illinois, American Academy of Sleep Medicine, 176(2012):7, 2012

    work page 2012

  4. [4]

    Interpretable feature-based machine learning for automatic sleep detection using photoplethysmography.npj Biosensing, 2(1):24, 2025

    Karmen Markov, Mohamed Elgendi, Vera Birrer, and Carlo Menon. Interpretable feature-based machine learning for automatic sleep detection using photoplethysmography.npj Biosensing, 2(1):24, 2025

    work page 2025

  5. [5]

    Bing Zhai, Ignacio Perez-Pozuelo, Emma A. D. Clifton, Joao Palotti, and Yu Guan. Making sense of sleep: Multimodal sleep stage classification in a large, diverse population using movement and cardiac sensing.Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., 4(2), June 2020

    work page 2020

  6. [6]

    Garcia-Gomez, Shahrad Taheri, Yu Guan, and Luis Fernandez-Luque

    Ignacio Perez-Pozuelo, Bing Zhai, Joao Palotti, Raghvendra Mall, Michaël Aupetit, Juan M. Garcia-Gomez, Shahrad Taheri, Yu Guan, and Luis Fernandez-Luque. The future of sleep health: a data-driven revolution in sleep science and medicine.npj Digital Medicine, 3(1):42, 2020

    work page 2020

  7. [7]

    Optimizing Photoplethysmography-Based Sleep Staging Models by Leveraging Temporal Context for Wearable Devices Applications

    Joseph AP Quino, Diego AC Cardenas, Marcelo AF Toledo, Felipe M Dias, Estela Ribeiro, José E Krieger, and Marco A Gutierrez. Optimizing photoplethysmography-based sleep staging models by leveraging temporal context for wearable devices applications.arXiv preprint arXiv:2410.00693, 2024

    work page Pith review arXiv 2024

  8. [8]

    Seiichi Morokuma, Toshinari Hayashi, Masatomo Kanegae, Yoshihiko Mizukami, Shinji Asano, Ichiro Kimura, Yuji Tateizumi, Hitoshi Ueno, Subaru Ikeda, and Kyuichi Niizeki. Deep learning-based sleep stage classification with cardiorespiratory and body movement activities in individuals with suspected sleep disorders.Scientific reports, 13(1):17730, 2023

    work page 2023

Show all 46 references
  1. [9]

    On improving ppg-based sleep staging: A pilot study

    Jiawei Wang, Yu Guan, Chen Chen, Ligang Zhou, Laurence T Yang, and Sai Gu. On improving ppg-based sleep staging: A pilot study. InCompanion of the 2025 ACM International Joint Conference on Pervasive and Ubiquitous Computing, pages 1640–1644, 2025

    2025

  2. [10]

    Performance of a convolutional neural network derived from ppg signal in classifying sleep stages.IEEE Transactions on Biomedical Engineering, 70(6):1717–1728, 2022

    Ahsan Habib, Mohammod Abdul Motin, Thomas Penzel, Marimuthu Palaniswami, John Yearwood, and Chandan Karmakar. Performance of a convolutional neural network derived from ppg signal in classifying sleep stages.IEEE Transactions on Biomedical Engineering, 70(6):1717–1728, 2022

    2022

  3. [11]

    Photoplethysmography—new applications for an old technology: a sleep technology review.Journal of Clinical Sleep Medicine, 19(1):189–195, 2023

    Scott Ryals, Ambrose Chiang, Sharon Schutte-Rodin, Arvind Chandrakantan, Nitun Verma, Steven Holfinger, Fariha Abbasi-Feinberg, Anuja Bandyopadhyay, Kelly Baron, Sumit Bhargava, et al. Photoplethysmography—new applications for an old technology: a sleep technology review.Journ...

    2023

  4. [12]

    Autonomic activity during human sleep as a function of time and sleep stage

    John Trinder, Jan Kleiman, Melinda Carrington, Simon Smith, Sibilah Breen, Nellie Tan, and Young Kim. Autonomic activity during human sleep as a function of time and sleep stage. Journal of sleep research, 10(4):253–264, 2001

    2001

  5. [13]

    Marek Malik. Heart rate variability: Standards of measurement, physiological interpretation, and clinical use: Task force of the european society of cardiology and the north american society for pacing and electrophysiology.Annals of Noninvasive Electrocardiology, 1(2):151–181, 1996

    1996

  6. [14]

    Sleepppg-net2: Deep learning generalization for sleep staging from photoplethysmography.arXiv preprint arXiv:2404.06869, 2024

    Shirel Attia, Revital Shani Hershkovich, Alissa Tabakhov, Angeleene Ang, Sharon Haimov, Riva Tauman, and Joachim A Behar. Sleepppg-net2: Deep learning generalization for sleep staging from photoplethysmography.arXiv preprint arXiv:2404.06869, 2024

    2024 arXiv

  7. [15]

    Sleepppg-net: A deep learning algorithm for robust sleep staging from continuous photoplethysmography.IEEE Journal of Biomedical and Health Informatics, 27(2):924–932, 2022

    Kevin Kotzen, Peter H Charlton, Sharon Salabi, Lea Amar, Amir Landesberg, and Joachim A Behar. Sleepppg-net: A deep learning algorithm for robust sleep staging from continuous photoplethysmography.IEEE Journal of Biomedical and Health Informatics, 27(2):924–932, 2022

    2022

  8. [16]

    A computation- ally efficient algorithm for wearable sleep staging in clinical populations.Scientific Reports, 13(1):9182, 2023

    Pedro Fonseca, Marco Ross, Andreas Cerny, Peter Anderer, Fokke van Meulen, Hennie Janssen, Angelique Pijpers, Sylvie Dujardin, Pauline van Hirtum, Merel van Gilst, et al. A computation- ally efficient algorithm for wearable sleep staging in clinical populations.Scientific Repo...

    2023

  9. [17]

    Assessment of obstructive sleep apnea-related sleep fragmentation utilizing deep learning-based sleep staging from photoplethysmography.Sleep, 44(10):zsab142, 2021

    Riku Huttunen, Timo Leppänen, Brett Duce, Arie Oksenberg, Sami Myllymaa, Juha Töyräs, and Henri Korkalainen. Assessment of obstructive sleep apnea-related sleep fragmentation utilizing deep learning-based sleep staging from photoplethysmography.Sleep, 44(10):zsab142, 2021

    2021

  10. [18]

    Sleep stage classification from heart-rate variability using long short-term memory neural networks.Scientific reports, 9(1):14149, 2019

    Mustafa Radha, Pedro Fonseca, Arnaud Moreau, Marco Ross, Andreas Cerny, Peter Anderer, Xi Long, and Ronald M Aarts. Sleep stage classification from heart-rate variability using long short-term memory neural networks.Scientific reports, 9(1):14149, 2019

    2019

  11. [19]

    Sleep stage classification with ecg and respiratory effort.Physiological measurement, 36(10):2027–2040, 2015

    Pedro Fonseca, Xi Long, Mustafa Radha, Reinder Haakma, Ronald M Aarts, and Jérôme Rolink. Sleep stage classification with ecg and respiratory effort.Physiological measurement, 36(10):2027–2040, 2015

    2027

  12. [20]

    A deep learning model for automated sleep stages classification using psg signals.International journal of environmental research and public health, 16(4):599, 2019

    Ozal Yildirim, Ulas Baran Baloglu, and U Rajendra Acharya. A deep learning model for automated sleep stages classification using psg signals.International journal of environmental research and public health, 16(4):599, 2019

    2019

  13. [21]

    Stanislas Chambon, Mathieu N Galtier, Pierrick J Arnal, Gilles Wainrib, and Alexandre Gram- fort. A deep learning architecture for temporal sleep stage classification using multivariate and multimodal time series.IEEE Transactions on Neural Systems and Rehabilitation Engineeri...

    2018

  14. [22]

    A convolutional neural network for sleep stage scoring from raw single-channel eeg.Biomedical Signal Processing and Control, 42:107–114, 2018

    Arnaud Sors, Stéphane Bonnet, Sébastien Mirek, Laurent Vercueil, and Jean-François Payen. A convolutional neural network for sleep stage scoring from raw single-channel eeg.Biomedical Signal Processing and Control, 42:107–114, 2018

    2018

  15. [23]

    A multi-class automatic sleep staging method based on photoplethysmography signals.Entropy, 23(1):116, 2021

    Xiangfa Zhao and Guobing Sun. A multi-class automatic sleep staging method based on photoplethysmography signals.Entropy, 23(1):116, 2021

    2021

  16. [24]

    Emadeldeen Eldele, Zhenghua Chen, Chengyu Liu, Min Wu, Chee-Keong Kwoh, Xiaoli Li, and Cuntai Guan. An attention-based deep learning approach for sleep stage classification with single-channel eeg.IEEE Transactions on Neural Systems and Rehabilitation Engineering, 29:809–818, 2021

    2021

  17. [25]

    L-seqsleepnet: Whole-cycle long sequence modeling for automatic sleep staging.IEEE Journal of Biomedical and Health Informatics, 27(10):4748–4757, 2023

    Huy Phan, Kristian P Lorenzen, Elisabeth Heremans, Oliver Y Chén, Minh C Tran, Philipp Koch, Alfred Mertins, Mathias Baumert, Kaare B Mikkelsen, and Maarten De V os. L-seqsleepnet: Whole-cycle long sequence modeling for automatic sleep staging.IEEE Journal of Biomedical and He...

    2023

  18. [26]

    Deepsleepnet: A model for automatic sleep stage scoring based on raw single-channel eeg.IEEE transactions on neural systems and rehabilitation engineering, 25(11):1998–2008, 2017

    Akara Supratak, Hao Dong, Chao Wu, and Yike Guo. Deepsleepnet: A model for automatic sleep stage scoring based on raw single-channel eeg.IEEE transactions on neural systems and rehabilitation engineering, 25(11):1998–2008, 2017

    1998

  19. [27]

    Towards long-term sleep staging via wearable reflective photoplethysmography.SLEEPJ, page zsaf246, 2025

    Loris Constantin, Christian M Horvath, Florent Baty, Clémentine Aguet, Jérôme Van Zaen, Alia Lemkaddem, Loïc Jeanningros, Martin Proença, Xiaoli Yang, Kurt De Jaegere, et al. Towards long-term sleep staging via wearable reflective photoplethysmography.SLEEPJ, page zsaf246, 2025

    2025

  20. [28]

    Deep learning en- ables sleep staging from photoplethysmogram for patients with suspected sleep apnea.Sleep, 43(11):zsaa098, 2020

    Henri Korkalainen, Juhani Aakko, Brett Duce, Samu Kainulainen, Akseli Leino, Sami Nikko- nen, Isaac O Afara, Sami Myllymaa, Juha Töyräs, and Timo Leppänen. Deep learning en- ables sleep staging from photoplethysmogram for patients with suspected sleep apnea.Sleep, 43(11):zsaa098, 2020

    2020

  21. [29]

    Getting more from less: Transfer learning improves sleep stage decoding accuracy in peripheral wearable devices.arXiv preprint arXiv:2506.00730, 2025

    William G Coon, Diego Luna, Akshita Panagrahi, Matthew Reid, and Mattson Ogg. Getting more from less: Transfer learning improves sleep stage decoding accuracy in peripheral wearable devices.arXiv preprint arXiv:2506.00730, 2025

    2025 arXiv

  22. [30]

    Optimal detection of changepoints with a linear computational cost.Journal of the American Statistical Association, 107(500):1590–1598, 2012

    Rebecca Killick, Paul Fearnhead, and Idris A Eckley. Optimal detection of changepoints with a linear computational cost.Journal of the American Statistical Association, 107(500):1590–1598, 2012

    2012

  23. [31]

    Bayesian online changepoint detection.arXiv preprint arXiv:0710.3742, 2007

    Ryan Prescott Adams and David JC MacKay. Bayesian online changepoint detection.arXiv preprint arXiv:0710.3742, 2007

    2007 arXiv

  24. [32]

    Hidden semi-markov models.Artificial intelligence, 174(2):215–243, 2010

    Shun-Zheng Yu. Hidden semi-markov models.Artificial intelligence, 174(2):215–243, 2010

    2010

  25. [33]

    Activity recognition from user-annotated acceleration data

    Ling Bao and Stephen S Intille. Activity recognition from user-annotated acceleration data. In International conference on pervasive computing, pages 1–17. Springer, 2004

    2004

  26. [34]

    A tutorial on human activity recognition using body-worn inertial sensors.ACM Computing Surveys (CSUR), 46(3):1–33, 2014

    Andreas Bulling, Ulf Blanke, and Bernt Schiele. A tutorial on human activity recognition using body-worn inertial sensors.ACM Computing Surveys (CSUR), 46(3):1–33, 2014

    2014

  27. [35]

    Moddrop: adaptive multi- modal gesture recognition.IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(8):1692–1706, 2015

    Natalia Neverova, Christian Wolf, Graham Taylor, and Florian Nebout. Moddrop: adaptive multi- modal gesture recognition.IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(8):1692–1706, 2015

    2015

  28. [36]

    Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network.Nature medicine, 25(1):65–69, 2019

    Awni Y Hannun, Pranav Rajpurkar, Masoumeh Haghpanahi, Geoffrey H Tison, Codie Bourn, Mintu P Turakhia, and Andrew Y Ng. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network.Nature medicine, 25(1):65–69, 2019

    2019

  29. [37]

    Sleepvst: Sleep staging from near-infrared video signals using pre-trained transformers

    Jonathan F Carter, João Jorge, Oliver Gibson, and Lionel Tarassenko. Sleepvst: Sleep staging from near-infrared video signals using pre-trained transformers. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 12479–12489, 2024

    2024

  30. [38]

    Aasm scoring manual updates for 2017 (version 2.4), 2017

    Richard B Berry, Rita Brooks, Charlene Gamaldo, Susan M Harding, Robin M Lloyd, Stuart F Quan, Matthew T Troester, and Bradley V Vaughn. Aasm scoring manual updates for 2017 (version 2.4), 2017

    2017

  31. [39]

    Photoplethysmography and its application in clinical physiological measurement

    John Allen. Photoplethysmography and its application in clinical physiological measurement. Physiological measurement, 28(3):R1–R39, 2007

    2007

  32. [40]

    Spindle frequency activity in the sleep eeg: individual differences and topographical distribution.Electroen- cephalography and clinical neurophysiology, 103(5):535–542, 1997

    Esther Werth, Peter Achermann, Derk-Jan Dijk, and Alexander A Borbély. Spindle frequency activity in the sleep eeg: individual differences and topographical distribution.Electroen- cephalography and clinical neurophysiology, 103(5):535–542, 1997

    1997

  33. [41]

    Racial/ethnic differences in sleep disturbances: the multi-ethnic study of atherosclerosis (mesa).Sleep, 38(6):877–888, 2015

    Xiaoli Chen, Rui Wang, Phyllis Zee, Pamela L Lutsey, Sogol Javaheri, Carmela Alcántara, Chandra L Jackson, Michelle A Williams, and Susan Redline. Racial/ethnic differences in sleep disturbances: the multi-ethnic study of atherosclerosis (mesa).Sleep, 38(6):877–888, 2015

    2015

  34. [42]

    Normal human sleep: an overview.Principles and practice of sleep medicine, 4(1):13–23, 2005

    Mary A Carskadon, William C Dement, et al. Normal human sleep: an overview.Principles and practice of sleep medicine, 4(1):13–23, 2005. 12

    2005

  35. [43]

    The familial aggregation of obstructive sleep apnea.American journal of respiratory and critical care medicine, 151(3):682–687, 1995

    Susan Redline, Peter V Tishler, Tor D Tosteson, John Williamson, Kenneth Kump, Ilene Browner, Veronica Ferrette, and Patrick Krejci. The familial aggregation of obstructive sleep apnea.American journal of respiratory and critical care medicine, 151(3):682–687, 1995

    1995

  36. [44]

    Sleepppg-net2: Deep learning generalization for sleep staging from photoplethysmography.Physiological Measurement, 46(12):125001, 2025

    Shirel Attia, Revital Shani Hershkovich, Alissa Tabakhov, Angeleene Ang, Arie Oksenberg, Riva Tauman, and Joachim A Behar. Sleepppg-net2: Deep learning generalization for sleep staging from photoplethysmography.Physiological Measurement, 46(12):125001, 2025

    2025

  37. [45]

    Interrater reliability for sleep scoring according to the rechtschaffen & kales and the new aasm standard.Journal of sleep research, 18(1):74–84, 2009

    Heidi Danker-Hopfe, Peter Anderer, Josef Zeitlhofer, Marion Boeck, Hans Dorn, Georg Gruber, Esther Heller, Erna Loretz, Doris Moser, Silvia Parapatics, et al. Interrater reliability for sleep scoring according to the rechtschaffen & kales and the new aasm standard.Journal of s...

    2009

  38. [46]

    Sympathetic-nerve activity during sleep in normal subjects.New England Journal of Medicine, 328(5):303–307, 1993

    Virend K Somers et al. Sympathetic-nerve activity during sleep in normal subjects.New England Journal of Medicine, 328(5):303–307, 1993. A Implementation Details, Resources, and Human-Subjects Notes A.1 Training Details To keep the comparison focused on supervision rather than...

    1993

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.