In the dimly lit sleep labs at Mount Sinai, machines quietly record the brain waves, breathing patterns, and heart rates of sleeping patients. Now, a newly developed artificial intelligence system can process this data with unprecedented comprehensiveness—analyzing entire nights of sleep rather than the traditional fragmented approach.
Researchers at the Icahn School of Medicine have unveiled an AI tool that has already processed a staggering 1,011,192 hours of human sleep, making it one of the largest sleep analysis studies ever conducted. The findings were published in the journal Sleep on March 13.
The model, named “patch foundational transformer for sleep” (PFTSleep), represents a significant departure from conventional sleep analysis methods. Traditional approaches typically involve human experts manually scoring short segments of sleep data or using AI models limited to analyzing brief intervals.
“This is a step forward in AI-assisted sleep analysis and interpretation,” says Benjamin Fox, first author and PhD candidate at the Icahn School of Medicine at Mount Sinai. “By leveraging AI in this way, we can learn relevant clinical features directly from sleep study signal data and use them for sleep scoring and, in the future, other clinical applications such as detecting sleep apnea or assessing health risks linked to sleep quality.”
Built on the same transformer architecture powering language models like ChatGPT, the system analyzes brain waves, muscle activity, heart rate, and breathing patterns throughout an entire night’s rest. This comprehensive approach allows it to identify patterns and relationships that might be missed when examining only isolated 30-second segments.
The technology arrives at a time when sleep disorders affect nearly a third of Americans, with many cases going undiagnosed due to the complex and time-consuming nature of sleep analysis. Sleep specialists spend hours reviewing polysomnograms—recordings that capture multiple physiological signals during sleep—to diagnose conditions like sleep apnea or insomnia.
Dr. Ankit Parekh, co-senior corresponding author and Assistant Professor of Medicine at the Icahn School of Medicine, sees broader applications on the horizon. “Our findings suggest that AI could transform how we study and understand sleep,” he explains. “Our next goal is to refine the technology for clinical applications, such as identifying sleep-related health risks more efficiently.”
What sets this model apart is its training methodology. Rather than relying solely on human-labeled data, the researchers employed “self-supervision”—allowing the AI to identify relevant patterns independently. This approach helps the system learn directly from physiological signals, potentially uncovering subtle relationships that human experts might overlook.
The team trained PFTSleep using three major sleep study databases: the Sleep Heart Health Study, Wisconsin Sleep Cohort, and Osteoporotic Fractures in Men Study. They then validated its performance against independent datasets, achieving impressive accuracy ratings. The model’s Cohen’s Kappa scores—a measure of inter-rater reliability—ranged from 0.59 to 0.81 across different test sets, comparable to expert human analysis.
Sleep medicine has traditionally struggled with standardization issues. Different sleep centers and individual specialists may interpret the same sleep data differently, leading to inconsistent diagnoses. A system like PFTSleep could help address this problem by providing a more uniform analysis approach.
“By analyzing entire nights of sleep with greater consistency, we can uncover deeper insights into sleep health and its connection to overall well-being,” notes Dr. Girish N. Nadkarni, co-senior corresponding author and Chair of the Windreich Department of Artificial Intelligence and Human Health at the Icahn School of Medicine.
Dr. Nadkarni, who also serves as Director of the Hasso Plattner Institute for Digital Health, emphasizes that such AI-driven approaches could revolutionize sleep research, though he cautions that the technology is meant to augment rather than replace clinical expertise.
The implications extend beyond sleep disorders. Growing evidence links sleep quality to cardiovascular health, cognitive function, and immune system performance. More efficient sleep analysis could help researchers better understand these connections and potentially identify early warning signs of health issues.
For the average person struggling with sleep problems, tools like PFTSleep might eventually lead to more accessible sleep assessments. Current polysomnography typically requires overnight stays in specialized facilities, making it inconvenient and costly. AI systems that can effectively analyze sleep data could potentially support home-based alternatives in the future.
The researchers are already looking ahead to expanding the model’s capabilities. Beyond classifying sleep stages, they hope to develop systems that can detect specific sleep disorders and even predict health outcomes based on sleep patterns.
As AI continues to permeate healthcare, this advancement in sleep analysis demonstrates how emerging technologies can address long-standing clinical challenges. Whether this leads to better diagnoses, more personalized treatments, or simply a better night’s sleep remains to be seen—but for now, while you sleep, AI is learning to watch over you.
If you found this piece useful, please consider supporting our work with a small, one-time or monthly donation. Your contribution enables us to continue bringing you accurate, thought-provoking science and medical news that you can trust. Independent reporting takes time, effort, and resources, and your support makes it possible for us to keep exploring the stories that matter to you. Together, we can ensure that important discoveries and developments reach the people who need them most.
