Researchers at the Icahn School of Medicine have developed a powerful AI tool based on the same transformer architecture used by large language models like ChatGPT to process an entire night of sleep. To date, it is one of the largest studies, analyzing 1,011,192 hours of sleep. The model, called the Patch-Foundational-Transformer for Sleep (PFTSleep), analyzes brain waves, muscle activity, heart rate, and breathing patterns to classify stages of sleep more effectively than conventional methods, optimize sleep analysis, reduce variability, and support future clinical tools for detecting sleep disorders and other health risks. Details of their findings have been published in the online edition of the journal “Sleep”.

Potential Areas Where AI Can be Applied

Artificial intelligence can make a significant contribution to the field of sleep medicine. AI in the field of sleep has potential applications in three key areas:

• Clinical applications: In the clinical setting, AI-driven technologies provide comprehensive data analysis, sophisticated pattern recognition, and automation in diagnosis while simultaneously addressing chronic issues such as sleep-disordered respiratory disorders. Despite modest beginnings, the use of AI can lead to improvements in efficiency and patient access, which can help reduce burnout among healthcare professionals.
• Lifestyle management: The integration of AI also offers clear benefits for lifestyle management through the use of consumer sleep technology. These devices come in a variety of forms, such as fitness trackers, smartphone apps, and smart rings, and help to improve sleep health by tracking, evaluating, and improving sleep. Wearable sleep technology and data-driven lifestyle recommendations can empower patients to take an active role in managing their health, as evidenced by a recent AASM survey in which 68% of adults who have used a sleep tracker reported changing behavior based on insights gained. However, as these AI-driven applications become more and more intuitive, ongoing dialog between patients and clinicians about the potential and limitations of these innovations remains critical.
• Population health: Beyond the individual care setting, AI technology is opening up a new approach to public health as it relates to sleep. AI has the exciting potential to synthesize environmental, behavioral, and physiological data, contributing to informed population-level interventions and addressing existing health care gaps.

Novel Technique Optimizes Sleep Scoring and Supports Future Clinical Tools for Detecting Sleep Disorders and Other Health Risks

Current sleep scoring often relies on human experts who manually evaluate short segments of sleep data, or on AI models that are unable to analyze an entire patient’s night. This new approach, developed based on thousands of sleep recordings, provides a more comprehensive view. Trained on full-length sleep data, the model can detect sleep patterns throughout the night and across different populations and settings, providing a standardized and scalable method for sleep research and clinical use, the researchers say. “This is a step forward in AI-powered sleep scoring and interpretation,” says first author Benjamin Fox, a PhD student at the Icahn School of Medicine at Mount Sinai in the Training Program in Artificial Intelligence and Emerging Technologies. Using AI in this way can directly infer relevant clinical features from sleep study signal data for use in sleep scoring and, in the future, other clinical applications such as sleep apnea detection or health risk assessment related to sleep quality.

The model was trained on a large dataset of sleep studies (polysomnograms), which measure key physiological signals including brain activity, muscle tone, heart rate and breathing patterns. Unlike conventional AI models that analyze only short 30-second segments, this new model considers the entire night of sleep, capturing more detailed and nuanced patterns. Furthermore, the model is trained through a method known as self-supervision, which helps learn relevant clinical features from physiological signals without the need for human labeling of outcomes. “Our findings suggest that AI could transform the way we screen and understand sleep,” says co-senior corresponding author Ankit Parekh, PhD, assistant professor of medicine (pulmonary, critical care and sleep medicine) at the Icahn School of Medicine at Mount Sinai and director of the Sleep and Circadian Analysis Group at Mount Sinai. ”

The researchers’ next goal is to refine the technology for clinical applications, for example, to more efficiently identify sleep-related health risks. The researchers emphasize that while this AI tool is promising, it cannot replace clinical expertise. Instead, it is designed to serve as a powerful aid for sleep specialists and to help speed up and standardize sleep analysis. Next, the team plans to direct their research toward expanding capabilities beyond sleep stage classification to sleep disorder detection and health outcome prediction.