Researchers at the Icahn Medicine School have developed a powerful AI tool, built on the same transformer architecture used by large language models as a chatgpt, to treat the whole night of sleep. To date, this is one of the largest studies, analyzing 1,011 192 hours of sleep. Details on their results were reported on March 13 Online number of the newspaper Sleep [https://doi.org/10.1093/sleep/zsaf061].
The model, called the fundamental transformer of patch for sleep (PFTSLEEP), analyzes brain waves, muscle activity, heart rate and respiratory patterns to classify sleep stages more effectively than traditional methods, rationalization of sleep analysis, reduction in variability and support for future clinical tools to detect sleep disorders and other risks for health.
Current sleep analysis is often based on human experts manually marking short sleep data segments or using AI models that are not able to analyze a patient’s sleep night. This new approach, developed using thousands of sleep recordings, has a more complete view. By training on full sleep data, the model can recognize sleep models throughout the night and in different populations and parameters, offering a standardized and scalable method for sleep and clinical use, explain the researchers.
This is a step forward in the analysis and interpretation of sleep assisted by AI. By taking advantage of AI in this way, we can learn the relevant clinical characteristics directly from sleeping signal data and use it for sleep rating and, in the future, other clinical applications such as sleep apnea detection or health risk assessment related to sleep quality. “”
Benjamin Fox, first author, doctoral candidate at Icahn School of Medicine in Mount Sinai in the formation area of artificial intelligence and emerging technologies
The model was built using a large set of sleep studies (polysomnograms) which measure key physiological signals, including brain activity, muscle tone, heart rate and breathing models. Unlike traditional AI models, which only analyze short segments and 30 seconds, this new model considers the whole sleep night, capturing more detailed and nuanced models. In addition, the model is formed via a method known as self-operation, which helps learn the relevant clinical characteristics from physiological signals without using human marked results.
“Our results suggest that AI could transform the way we study and understand sleep,” explains the corresponding author Co-Senior, Ankit Parekh, PHD, assistant professor of medicine (pulmonary, intensive care and sleep medicine) at the Icahn Medicine School at Mount Sinai and director of the Sleep and Circadian Group of the Mount Sinai. “Our next objective is to refine clinical applications technology, such as health risk identification related to sleep more effectively.”
Researchers point out that this promising AI tool would not replace clinical expertise. Instead, this would serve as a powerful help for sleep specialists, helping to speed up and normalize sleep analysis. Then, the team’s search aims to extend its capacities beyond the classification of the sleep stadium to detect sleep disorders and predict health results.
“This AI-based approach has the potential to revolutionize research on sleep,” said the corresponding author Co-Senior, Girish N. Nadkarni, MD, MPH, president of the Winduch Department of Artificial Intelligence and Human Health at Icahn School of Medicine, Director of Hasso Plattner Institute for Digital HEALTH DIGITAL,, and Irene and Dr Arthur Mr. Fishberg professor of medicine. Dr. Nadkarni is also the inaugural chief of the data division focused on data and digital and co -director of the Mount Sinai Clinical Center. “By analyzing whole nights of sleep with greater consistency, we can discover more in-depth information on sleep health and its link with general well-being.”
The paper is titled “A fundamental transformer taking advantage of the full night, data from the study of multichannel sleep with precision sleep stages.“”
The authors of the study, listed in the journal, are Benjamin Fox, Joy Jiang, Sajila Wickramaratne, Patricia Kovatch, Mayte Suarez-Farinas, Neomi A. Shah, Ankit Parekh and Girish N. Nadkarni.
Source:
Journal reference:
Fox, B., et al. (2025). A fundamental transformer taking advantage of the full night, data from the study of multichannel sleep classified with precision the sleep stages. Sleep. Doi.org/10.1093/sleep/zsaf061.
