How AI detects over 100 health conditions from just one night of sleep: Here's every detail you should know

A new artificial intelligence (AI) model named SleepFM can flag a person's risk of developing over 100 health conditions based on how they sleep.

SleepFM, a large language model, is developed by a team of researchers at California's Stanford University.

SleepFM is a large language model (LLM) developed that’s capable of reading a user’s brain activity, heart rate, respiratory signals, and leg movements while they’re sleeping to assess the risk of disease.

The study was published in the journal Nature, showing that researchers have trained the AI model SleepFM, using over 580,000 hours of sleep data obtained from 65,000 patients between the period from 1999 and 2004.

The data has been accessed from sleep clinics, medical facilities that examine sleep patterns overnight, and were divided into five-second increments, which functioned like words for LMMs to train on.

Study co-author James Zou said, “SleepFM is essentially learning the language of sleep.”

To enable disease prediction, the researcher trained SleepFM using both the sleep data and the patient’s clinical health files.

With an accuracy rate of at least 80%, SleepFM (LMM) correctly predicted the onset of conditions like Parkinson’s, dementia, Alzheimer’s, hypertensive heart disease, cardiovascular disease, and prostate and breast cancer.

Despite an overall accuracy rate of 80%, the AI model, SleepFM, was less accurate for chronic kidney disease, stroke, and arrhythmia, detecting these conditions in at least 785 of cases.

The researchers noted that their study only involved participation from sleep clinics that already had health concerns, meaning the results may not reflect the AI’s ability to detect disease among the masses.