Neurology, development and sleep
Software that flags seizures on EEG, scores sleep studies and aids autism diagnosis in young children, producing results for neurologists, sleep physicians and pediatric clinicians to review.
Category overviewHide category overview
These tools analyze EEG, sleep study signals or developmental assessments: limited-montage EEG with automated seizure burden estimates, seizure detection software for conventional EEG, automated scoring of sleep studies and an aid for diagnosing autism in young children. FDA cleared the EEG seizure detection software and the sleep study scoring software reviewed here as automatic event detection software under 21 CFR 882.1400; the labeling of Ceribell's status epilepticus software says it does not substitute for a qualified clinician's review of the EEG for other abnormal patterns. Autism diagnosis aids got their own device classification when FDA granted Cognoa's De Novo request on June 2, 2021.
Most of the evidence measures agreement with expert review, and experts themselves disagree: in 120 epilepsy monitoring unit recordings, pairwise sensitivities among three experts ranged from 72.5% to 84.9%. In 7,924 critical care EEG records, one widely used detector had a record-level positive predictive value of 0.08. The American Academy of Sleep Medicine's artificial intelligence committee has said caution must be applied when using AI. Open questions include whether faster seizure detection improves neurological outcomes, which no randomized trial of these products has tested.
Canvas Dx
Prescription software that combines a caregiver questionnaire, home videos and a clinician questionnaire to aid autism diagnosis in young children.
Full reviewHide full review: Canvas Dx
Canvas Dx combines a caregiver questionnaire, analysis of two short home videos and a health care provider questionnaire in a machine learning algorithm that returns autism positive, autism negative or indeterminate, according to its pivotal study. FDA labeling says it is not a stand-alone diagnostic but an adjunct to the diagnostic process, according to the De Novo summary.
Why this grade. A company-coauthored prospective study in primary care measured use in care: time from clinical concern to autism diagnosis averaged 39.22 days with the device, against waits of 180 to 264 days at nearby specialist centers, without a concurrent control group. No randomized trial was found; the pivotal study measured accuracy.
The evidence
In the pivotal study, a double-blinded prospective study at multiple sites, 425 children aged 18 to 72 months with developmental concerns completed device and specialist assessments. Positive predictive value was 80.8% and negative predictive value 98.3%, but only 31.8% received a determinate result; among them, sensitivity was 98.4% and specificity 78.9%. Without the indeterminate output, sensitivity would fall to 51.6% and specificity to 18.5%. Specialists found that 91% of children without a result had one or more complex neurodevelopmental disorders.
In the ECHO Autism study, the device gave a determinate result for 52.5% of children, all consistent with the final clinical determination; 93% of responding caregivers were satisfied, and clinicians said it did not remove the need for structured observation in every case. In a company analysis of 254 prescriptions, 63.0% received a determinate result, with positive predictive value of 92.4% and negative predictive value of 97.6%.
- Used by
- Clinicians in the ECHO Autism primary care program used it in a prospective study of 80 children; Cognoa analyzed its first 254 prescriptions after authorization.
- FDA
- FDA granted the De Novo request for the Cognoa ASD Diagnosis Aid on June 2, 2021 (DEN200069), for children 18 to 72 months old at risk for developmental delay. Canvas Dx was cleared through K243558 on April 11, 2025, adding a predetermined change control plan; no clinical testing was performed for that submission. Original authorization was De Novo DEN200069 (2021) as the Cognoa ASD Diagnosis Aid; Canvas Dx was cleared through 510(k) K243558 (2025), adding a predetermined change control plan.
- Randomized trial
- None published
- Key study
- Scientific Reports, 2025, company analysis of the first 254 prescriptions after authorization
- Limits
- Every published study was funded or coauthored by Cognoa. The primary care study had no control group, compared its timeline with specialist waits rather than the same program without the device, and used clinicians trained in autism diagnosis, and most children in the pivotal study received no result.
- Cost
- A November 2024 health technology assessment prepared for the New York State Department of Health by Oregon Health & Science University's Center for Evidence-based Policy lists an average wholesale price of $2,280 per Canvas Dx evaluation. checked
- Specialties
- Family medicine, Pediatrics
Ceribell
A rapid-setup, limited-montage EEG headband with AI software that estimates seizure burden and flags electrographic status epilepticus.
Full reviewHide full review: Ceribell
Nurses, residents or technologists place a 10-electrode headband, and the Clarity algorithm estimates seizure burden as the percentage of time in seizure over rolling five-minute windows and issues alerts for possible status epilepticus, according to a 2026 validation study. Physicians can also listen to sonified EEG at the bedside, as in the DECIDE study.
Why this grade. Studies have measured effects on care without randomization: in DECIDE, physicians' seizure diagnoses in 164 intensive care patients became more accurate after bedside EEG, and a retrospective matched analysis found shorter intensive care stays than with conventional EEG. No randomized trial was found.
The evidence
In DECIDE, a prospective study at five academic hospitals sponsored by Ceribell, sensitivity of physicians' seizure diagnosis rose from 77.8% to 100% and specificity from 63.9% to 89% with the device; EEG began in a median of 5 minutes, while conventional EEG was delayed a median of 239 minutes. In a subanalysis of 283 intensive care patients at three academic centers, stay in intensive care was 3.9 versus 8.0 days with conventional EEG; the authors said causation needs randomized studies.
In the FDA submission, the software detected all 10 cases of status epilepticus among 350 adults, with 94% specificity. In AccuRASE, at three Yale hospitals, an earlier version at a burden above 10% had a negative predictive value of 99% but a positive predictive value of 11% for status epilepticus. On the same recordings, version 8 reached 100% sensitivity above 20% burden, and its built-in alerts had 86% sensitivity and 94% specificity; several authors had consulted for Ceribell.
- Used by
- A community hospital in Syracuse, N.Y., reported using it in 189 patients over two years, with two patients transferred for continuous EEG monitoring.
- FDA
- FDA cleared the Ceribell Pocket EEG Device through 510(k) K191301 on Sept. 11, 2019, and the Ceribell Status Epilepticus Monitor software, for diagnosing electrographic status epilepticus in adults, through K223504 on May 23, 2023; its labeling says it does not substitute for clinician review of the EEG for other abnormal patterns. A delirium monitor was cleared through K251936 on Dec. 8, 2025.
- Randomized trial
- None published
- Key study
- Neurocritical Care, 2024, retrospective matched analysis against conventional EEG, 283 intensive care patients at three academic centers
- Limits
- No randomized trial has been published, and the outcome studies are observational, retrospective or company-sponsored. The validation sets held few cases of status epilepticus (seven in AccuRASE), and the AccuRASE II authors noted that the reduced montage misses occasional seizures that full-montage EEG detects.
- Cost
- No list price is published. Ceribell describes the system as eligible for a Medicare new technology add-on payment of up to $913.90 per eligible patient, citing the fiscal 2024 inpatient payment rule. checked
- Specialties
- Pulmonary and critical care, Neurology
EnsoSleep
AI software that automatically scores sleep studies, staging sleep and detecting apneas, hypopneas, arousals and leg movements.
Full reviewHide full review: EnsoSleep
EnsoSleep analyzes physiological signals from sleep studies and automatically scores sleep stages, arousals, leg movements and obstructive apneas, central apneas and hypopneas; all scored events are subject to verification and editing by a qualified clinician, according to its FDA summary. EnsoData's Aurora software analyzes photoplethysmography from compatible pulse oximeters to aid in diagnosing sleep apnea, its FDA summary says.
Why this grade. The evidence is accuracy data: the K210034 summary compared its scoring of 100 adult and 47 pediatric archived studies with technologist consensus, and a 2026 study by company researchers validated EnsoSleep photoplethysmography scoring against polysomnography in 225 people. No study of its effect on care was found.
The evidence
In the FDA submission, reference scores came from a two-of-three consensus of registered sleep technologists. Overall sleep stage agreement was 86.6% in adults, against 78% for the predicate device, but 37.0% for stage N1. For classifying an apnea-hypopnea index of 15 or more, overall agreement was 95.0%. Positive agreement for individual events was lower: 66.3% for adult hypopneas and 45.5% for pediatric obstructive apneas.
In the 2026 validation, in which all authors listed EnsoData affiliations, simultaneous polysomnography and photoplethysmography were recorded for one night in 225 people at risk for obstructive sleep apnea; sensitivity was 90.9% and specificity 72.3% for sleep apnea, and the mean apnea-hypopnea index was 0.74 events an hour higher than on polysomnography. In the Aurora submission, across 158 adults, sensitivity was 92.6% and specificity 71.6% at an index of 5 or more.
- FDA
- FDA cleared EnsoSleep through 510(k) K210034 on June 16, 2021, for patients 13 and older, including adult home sleep tests, and EnsoData's photoplethysmography software Aurora through K231355 on Feb. 9, 2024.
- Randomized trial
- None published
- Key study
- FDA 510(k) summary K210034, 2021, comparison with technologist consensus scoring, 100 adult and 47 pediatric archived studies
- Limits
- All published evidence was generated by the company or for its FDA submissions, on retrospective or single-night data. No study measured scoring time saved, diagnostic decisions or patient outcomes, and agreement for individual respiratory events was modest.
- Cost
- No published price found. checked
- Specialties
- Pulmonary and critical care, Neurology
Persyst
EEG review software whose automated detectors mark possible seizures and spikes and display quantitative trends for clinicians.
Full reviewHide full review: Persyst
Persyst marks sections of adult EEG recordings that may correspond to electrographic seizures, to assist qualified practitioners in assessing the EEG, according to its FDA indication. FDA labeling says its seizure notifications cannot substitute for real-time monitoring of the EEG by a trained expert.
Why this grade. Studies of Persyst measure detection accuracy against human review, such as a Persyst-coauthored study that found its detector statistically noninferior to expert readers; a workflow study compared diagnoses in 50 recordings, and an epilepsy unit study described nurses' reliance on its alarm without a comparison group. None has measured an effect on patient care or outcomes.
Caution. In 7,924 critical care EEG records at the University of Pennsylvania, Persyst's record-level positive predictive value was 0.08, and in a manually reviewed cohort it detected 50% of individual seizures; the authors concluded that it cannot accurately identify records that contain seizures and that its value lies in triaging low-risk EEGs.
The evidence
In the Pennsylvania study, record-level sensitivity was 0.91 and negative predictive value 0.97 in the expanded cohort. In 120 epilepsy monitoring unit EEGs, Persyst 14 had 78.2% sensitivity and 1.0 false positive a day, meeting noninferiority criteria against three experts; two authors worked for Persyst. In 81 patients in Vienna, Persyst 13 detected 81% of seizures per patient with 0.9 false alarms an hour; in a follow-up of 80 patients, with two Persyst employees among the authors, Persyst 14 detected 67.3% to 71.1% with 2 to 4 false positives a day.
In critically ill children, Persyst 13 detected 74.4% of 379 seizures with 15.5 false positives a day. In an epilepsy monitoring unit, its alarm detected 87.2% of focal impaired awareness and bilateral tonic-clonic seizures, and nurses relied on it more than on push-button alarms for focal impaired awareness seizures. In 50 prolonged video-EEGs, sampled visual review plus Persyst 13 reached the same diagnoses as complete review.
- FDA
- FDA cleared Persyst 14 through 510(k) K182181 on Nov. 29, 2018, and Persyst 15 through K222002 on Dec. 30, 2022, and again through K252160 on Oct. 31, 2025, as automatic event detection software for EEG.
- Randomized trial
- None published
- Key study
- Seizure, 2019, epilepsy monitoring unit study of the seizure alarm in clinical use, no comparison group
- Limits
- The studies are retrospective or small prospective accuracy comparisons with differing reference standards and software versions, and false positive rates vary widely by setting. None measured time saved, treatment decisions or patient outcomes, and some were coauthored by Persyst employees.
- Cost
- No published price found. checked
- Specialties
- Neurology
Reviews of the published evidence, not medical advice or an endorsement of any product.