Neurology · Sleep medicine, automated polysomnography scoring
Onera SleepMap
Onera B.V.
SleepMap is the scoring half of a two-part home polysomnography system. Onera's patch-based recorder has two peer-reviewed papers from one 206-patient validation against in-laboratory polysomnography. The scoring software that the FDA cleared in March 2026 has one: a 98-night manufacturer validation in the 510(k) summary, in which the algorithm agreed with a three-scorer majority on 91.6 per cent of sleep-stage epochs but found only 57 per cent of hypopnoeas. It is a technician's first pass, cleared for a single recorder, with every event still owned by the reviewing clinician.
Performance Metrics
Clinical Evidence
The only evidence on the scoring algorithms themselves is the validation in the K253668 510(k) summary. Onera ran SleepMap over 98 polysomnography nights from 98 patients recorded with the Onera STS I: 72 home studies in the United States (August 2023 to August 2025) and 26 in-clinic studies in Germany (September 2022 to April 2024). Three US-certified sleep professionals scored every recording to AASM rules, and a two-of-three majority served as ground truth. Epochs where the scorers could not agree (3,282, or 3.2 per cent) were excluded from the staging analysis. This is a retrospective analysis of archived recordings, not a prospective trial, and it has not been peer reviewed. Sleep staging is the strong result. Overall agreement across the five stages was 91.6 per cent (95 per cent CI 90.7 to 92.4), with positive agreement of 78.9 per cent and negative agreement of 94.7 per cent, and 21.1 per cent of all stage calls were wrong. Wake and REM were the cleanest stages (positive agreement 82.5 and 82.7 per cent, false discovery rates 7.1 and 11.5 per cent). N3 had the highest positive agreement, 85.6 per cent, but 41.4 per cent of the epochs labelled N3 were something else. N2, half of all epochs, was found at 75.4 per cent with an 8.7 per cent false discovery rate. N1 was weakest: the algorithm caught 69.6 per cent of N1 epochs but 90.4 per cent of the epochs it labelled N1 were something else. Event detection is more uneven. Obstructive apnoeas were found at 84.5 per cent positive agreement and desaturations at 87.5 per cent. Hypopnoeas were found at 56.9 per cent (52.2 to 60.9) with a 59.8 per cent false discovery rate, and arousals at 59.8 per cent. Leg movements were found at 77.6 per cent but 77.8 per cent of flagged leg movements were false. Read as intended, the numbers describe a pre-scorer whose output a technician edits rather than accepts. The FDA table reports agreement, not sensitivity and specificity against a clinical diagnosis, and gives no apnoea-hypopnoea index (AHI) agreement or severity classification for SleepMap. Two peer-reviewed studies validate the recorder that SleepMap depends on, not the software. Viniol and colleagues (Journal of Clinical Sleep Medicine 2025) recorded 206 patients at seven German sites with the Onera STS and a conventional polysomnograph at the same time, scored by three blinded scorers: concordance for AHI was 0.94 and for total sleep time 0.87, with an average epoch kappa of 0.62. Woehrle and colleagues (ERJ Open Research 2026) report the respiratory outcomes of the same cohort: AHI correlation 0.90, hypopnoea index 0.68, and AUC of 0.8 or better at AHI thresholds of 5, 15 and 30. Both were funded by Onera and run through a contract research organisation; one author is an Onera contractor and shareholder. A US home-feasibility study (Uhles, Sleep and Breathing 2025) found 83 per cent of 41 mailed-out studies diagnostic on expert assessment, with human double scoring. None of the three used SleepMap for scoring, so they say nothing about the algorithm and are not counted as device evidence here. AurisMark reasoning: one FDA clearance and a public validation earn one mark. A second is withheld because the algorithm evidence is a single retrospective manufacturer validation, there is one jurisdiction six months after clearance, and no independent or peer-reviewed study of the scoring software exists.
| Study | Design | n | Sensitivity | Specificity | AUC | Published |
|---|---|---|---|---|---|---|
| Onera B.V. (SleepMap automated scoring validation, FDA 510(k) K253668) | RetrospectiveRetrospective | 98 | 78.9% (staging positive agreement, all stages) | 94.7% (staging negative agreement, all stages) | — | FDA 510(k) summary K253668, decision 8 March 2026. 98 Onera STS I nights (72 US home, 26 German clinic), 2-of-3 majority of three US-certified scorers as ground truth, 95% bootstrap CIs. Overall staging agreement 91.6%. Event positive agreement: obstructive apnoea 84.5%, hypopnoea 56.9%, desaturation 87.5%, arousal 59.8%, leg movement 77.6%, PLM 72.0%. Not peer reviewed; agreement metrics, not diagnostic sensitivity and specificity. |
Inside the algorithm
Editorial featureHow a night of patch recordings becomes a scored sleep study.
Five stages — from raw input to verdict — drawn from manufacturer documentation and the public regulatory record.
- INGEST
- NORMALISE
- DETECT
- LOCALISE
- VERDICT
Stage 01 · INGEST
One recorder, one file format.
SleepMap accepts a single input: the EDF file exported from an Onera STS I study. The four adhesive patches (forehead, chest, abdomen, lower leg) supply frontal EEG, EOG, masseter (jaw) EMG, ECG, forehead pulse oximetry, bio-impedance flow and effort, nasal pressure, position, activity, sound level and leg EMG. The clearance does not cover recordings from any other polysomnograph, which is the main difference from its predicate EnsoSleep.
Input
Type II home or in-lab PSG, one night
Inference
Cloud, on upload
Inside the Auris+ Listing
Five more sections complete this device’s Auris+ Listing.
Decision Ledger
ProPro unlocks a private, cross-vendor log of every case you read on this device — what the AI called, what you concluded, and a one-line reason if you overrode it. Ready for the EU AI Act's deployer logging obligations when they land in 2028.
Clinical Evidence Deep Dive
ProPro unlocks the structured clinical-evidence summary — study count, target patient population, and a tabular accuracy-metrics view drawn from peer-reviewed sources.
Peer-Reviewed Publications
ProPro unlocks the curated peer-reviewed publication list with PubMed cross-links — the citation backbone of every editorial verdict.
Post-Market & Regulatory Conditions
ProPro unlocks the post-market surveillance summary, recall record, and the conditions of approval that bound real-world use.
AI Algorithm Version History
ProPro unlocks the chronological record of algorithm version changes — what changed when, drawn from manufacturer changelogs and regulatory filings.
Regulatory Approvals
Safety Record
No recall, enforcement action or MAUDE adverse-event report naming Onera or SleepMap was found in the openFDA device databases, queried directly on 7 September 2026. The safety question for an autoscorer is under-detection that survives review. In the FDA validation the algorithm missed 43 per cent of hypopnoeas and 40 per cent of arousals, and it does not attempt central or mixed events, Cheyne-Stokes breathing or arrhythmias at all. A clinician who accepts the automated hypopnoea count without reviewing the flow and desaturation channels will under-score AHI in a patient whose disease is mainly hypopnoeic. The labelling makes clinician verification a condition of use.
Intended Use & Indications
Per the FDA indications for use (K253668): SleepMap is intended for use as an aid for physicians in the diagnosis of sleep and respiratory related disorders in adult patients (aged 18 and up) who have been prescribed a sleep study by their doctor. It is a software-only medical device used under the supervision of a clinician to analyze physiological signals and automatically score sleep study recordings, including the staging of sleep and the detection of arousals, leg movements, desaturations, obstructive apneas and obstructive hypopneas. Respiratory event subtypes (central and mixed apneas, central hypopneas), RERA, Cheyne Stokes breathing, snoring events and arrhythmia events are not automatically detected and must be manually marked. All automatically scored stages, detected events and physiological signals are subject to verification by a qualified clinician.