Researchers at Stanford University, along with their colleagues, have developed an innovative AI system that predicts the risk of developing numerous diseases by analyzing data from just one night of sleep.
By examining a complete set of physiological recordings from a single night of sleep, the SleepFM model forecasts the likelihood of , heart failure, stroke, and over a hundred other ailments.
What is this invention?
SleepFM is a foundational model, similar to , trained on a vast dataset. It draws from approximately 600,000 hours of sleep data collected from 65,000 individuals.
While ChatGPT learns from words and text, SleepFM learns from information about 5-second intervals of sleep.
“Large language models (LLMs) master our language by correlating words and text, whereas SleepFM essentially learns the language of sleep,” noted James Zhu, a biomedical data collection expert and co-author of the study.
The necessary array of information was gathered by the team across various clinics using a method known as polysomnography (PSG), which is considered the “gold standard” in sleep research. This method involves using sensors to track brain, heart, and respiratory activity during sleep, as well as movements of the legs and eyes.
“By studying sleep, we record an incredible number of signals,” says Emmanuel Mignot, a professor of sleep medicine at Stanford and one of the lead authors of the study.
The researchers tested SleepFM using a technique called contrastive learning, with one exception. During testing, the team excluded data such as heart rate or airflow for breathing, forcing SleepFM to extrapolate the missing information based on other biological data.
In this process, scientists compared polysomnography data with tens of thousands of treatment outcome reports from patients of various ages over a 25-year period. Analyzing information from medical records on 1,041 diseases, SleepFM accurately predicted 130 of them based on patients’ sleep data.
In its predictions, the SleepFM system outperformed established predictive models, according to Science Alert. The model achieved particular success in predicting Parkinson’s disease, myocardial infarction, stroke, chronic kidney disease, pregnancy complications, mental disorders, prostate cancer, breast cancer, and overall mortality. This further underscores the dangerous health consequences of poor sleep habits, the team noted.
The most reliable predictors of disease turned out to be physiological functions that had lost balance. “For example, a and a that hasn’t rested likely indicated problems,” explained Professor Mignot.
As the study demonstrated, the potential of artificial intelligence is incredibly valuable for medicine. This is primarily because AI can help save lives. In future scenarios, researchers plan to combine the capabilities of the SleepFM model with sleep monitoring devices to more closely examine people’s health in real-time.
The study’s findings were published in the journal Nature Medicine.
Photo: Unsplash
