SleepFM Sleep Data Analysis: The relationship between humans and disease is an age-old one. Previously, diseases were often detected much later. With the advancement of science, we can now learn about them much earlier.
Diagnosing sleep disorders
SleepFM Health Risk Prediction: Previously, you would only become aware of a disease when your condition had worsened significantly. But with the advancement of science, you can now determine the severity of a disease and what diseases you might develop in the future. Researchers have developed an artificial intelligence model that can predict a person's risk of 130 future diseases based on sleep data. This model has been named Sleep FM.
This model was developed by researchers from several institutions, including Stanford University in the US. It was trained using approximately 600,000 hours of sleep data collected from 65,000 people. The results of this research have been published in the medical journal Nature Medicine. Let us explain how this device works and how it can predict your future illnesses.
How does Sleep FM work?
Initially, this AI system was tested to identify common sleep-related problems, such as tracking different stages of sleep or predicting the severity of sleep apnea. Next, the sleep data was combined with patients' medical records to assess their risk of future illnesses. Researchers reported that the model was able to accurately predict 130 of the more than 1,000 illnesses listed in the health records.
Important health signs are hidden in sleep.
According to Emmanuel Mignot, professor of sleep medicine at Stanford University, "During sleep, a vast number of signals are recorded from the body. For eight hours, normal body activity is studied in such depth that the data becomes extremely rich."
What kind of data is captured?
Polysomnography was used to assess sleep, which is considered the most reliable method of sleep study. It uses sensors to:
As multiple signals are recorded, SleepFM interprets all these data streams simultaneously and analyzes their relationships.
A new way to train AI
The team used a technique called 'leave-one-out' contrastive learning to train the AI. This deliberately withholds one piece of data and challenges the AI to infer the missing information based on the remaining signals. This improves the model's understanding and accuracy.
Which diseases are best identified?
Research found that this AI is especially
It is quite robust in predicting many diseases. In many cases, its C-index score exceeded 0.8, indicating good predictive performance. According to the researchers, the diseases that Sleep FM was able to predict based on just one night's sleep data include:
Additionally, the model has been shown to be effective in predicting the risk of diseases like Parkinson's and developmental difficulties in children. Overall, this research suggests that sleep isn't just a way to relieve fatigue, but can also be a predictor of future health.
