SleepStageML software utilizes machine learning to automatically categorize sleep stages from EEG signals during polysomnography.

Key Points: 

  • The software was developed based on extensive data from hundreds of thousands of hours of sleep studies, showing performance on par with or superior to human experts in clinical validation tests.
  • It is the first sleep-focused medical device cleared by the FDA with a Predetermined Change Control Plan, allowing for continuous improvements to its machine-learning algorithm within the scope of its initial clearance.

The US Food and Drug Administration has granted 510(k) clearance to Beacon Biosignals’ SleepStageML, machine-learning software that automatically stages sleep from electroencephalogram (EEG) signals of clinical polysomnography (PSG) recordings to aid in the diagnosis and evaluation of sleep and sleep-related disorders. 

SleepStageML leverages a deep-learning model to score sleep stages. The model was trained on a massive dataset containing hundreds of thousands of hours of PSG recordings from both healthy individuals and patients with a diverse set of sleep disorders, neurologic disease, and psychiatric disease, acquired across numerous clinical sites, according to a release from Beacon Biosignal. 

Clinical validation testing demonstrated that SleepStageML performs as well or better than individual human experts as reported in literature.

According to a release from Beacon Biosignals, key benefits of SleepStageML include:

  • Automating the labor-intensive manual sleep staging process
  • Reducing subjective variability in scoring between human experts
  • Supporting faster PSG analysis turnaround time

“The advanced machine-learning algorithms powering SleepStageML reduce human scoring variability and increase precision of sleep measurements,” says Brandon Westover, MD, PhD, Landau Professor of Neurology at Harvard Medical School and co-founder of Beacon Biosignals, in a release. “This automated approach unlocks new clinical insights to drive forward therapy development for sleep and sleep-related disorders.”

SleepStageML is also the first medical device within the sleep space to be FDA-cleared with a Predetermined Change Control Plan, allowing Beacon to continuously improve the sleep staging machine-learning algorithm while still operating under the initial 510(k) clearance. The Predetermined Change Control Plan outlines strict validation testing that must be passed for any updates to the artificial intelligence/machine-learning model, ensuring improved performance compared to the originally cleared version. This allows Beacon to iteratively refine SleepStageML’s algorithms over time while verifying through comprehensive testing that the updated software meets safety and efficacy standards.

“SleepStageML’s approved [Predetermined Change Control Plan] is a game-changing development for the sleep field,” says Alexander Chan, PhD, vice president of analytics and machine learning at Beacon Biosignals, in a release. “With this regulatory pathway, we can provide even more accurate and robust sleep staging capabilities over time. This ability to iteratively enhance SleepStageML will be invaluable for generating insights to accelerate sleep therapy research and development.”

The clearance complements Beacon’s previous FDA clearance of the Dreem 3S wearable headband and integrated algorithms. 

“With FDA clearances for both SleepStageML and Dreem 3S headband, Beacon now provides an unparalleled capability to measure sleep physiology whether studies are conducted in-home or in-clinic,” says Jacob Donoghue, MD, PhD, CEO of Beacon Biosignals, in a release. 

Illustration 287962009 © Wuttichai Kaewklang |