Researchers say artificial intelligence can use brain recordings from a single night in a sleep lab to predict a person's risk of developing more than 100 health conditions.
The artificial intelligence model, known as SleepFM, has been trained with over half a million hours of sleep data, collected through polysomnography from around 65.000 participants.
Polysomnography is considered the gold standard for assessing overnight sleep. It uses various sensors to record brain activity, heart activity, breathing signals, body movements, eye movements, and other physiological data.
According to the researchers, this constitutes an "untapped goldmine of physiological data."
"We record an extraordinary number of signals when we study sleep," said study leader Dr. Emanuel Mignot of Stanford University School of Medicine in a statement, the Telegraph reports.
To harness the potential of this data, the researchers built an artificial intelligence model and trained it with 585.000 hours of polysomnographic data from patients whose sleep had been assessed at various sleep clinics.
Initially, the model was tested on standard sleep analysis tasks, such as classifying different sleep stages and diagnosing the severity of sleep apnea. SleepFM It proved to be as accurate, and in some cases even better, than the most advanced models currently in use.
Of the more than 1.000 disease categories analyzed, the model identified 130 conditions that could be predicted with reasonable accuracy from a patient's sleep data alone.
The researchers then compared polysomnography data from 35.000 adults and children treated at the Stanford Sleep Medicine Center between 1999 and 2024, with the same participants' long-term health outcomes, using their electronic health records.
Among the diseases that the model was able to predict include all-cause mortality, dementia, heart attack, heart failure, chronic kidney disease, stroke, and atrial fibrillation.
For some types of cancer, pregnancy complications, circulatory problems and mental disorders, the artificial intelligence model's predictions were accurate in more than 80% of cases, according to a report published in Nature Medicine.
However, researchers still do not fully understand what specific signals it analyzes. SleepFM when predicting a particular disease. They are working to clarify this, as well as to further improve the model's accuracy, perhaps by adding data from wearable health monitoring devices. /Telegraph/
