The artificial intelligence in question analyzes data about a person's condition during sleep, such as brain waves, heart rate, and breathing. This allows it to draw conclusions about the likelihood of developing various diseases. As noted by one of the study's authors, Stanford University associate professor James Zou, AI can predict disease risk many years before the first symptoms appear. The model, named SleepFM, was developed under the guidance of Rahul Tapa, a specialist in biomedical data.
From Sleep Signals to Disease Predictions
Polysomnography is a method of studying sleep that is usually conducted over one night and allows for monitoring the functioning of various body systems: the brain, heart, breathing, and muscles, as well as eye and limb movements. For training the SleepFM model, approximately 585,000 hours of recordings from 65,000 individuals who underwent examinations primarily at Stanford's sleep medicine center were used.
During the pre-training phase, the AI learned to analyze signals from the brain, heart, and breathing during sleep. Later, the model was refined to address tasks such as diagnosing apnea and determining sleep stages, achieving results comparable to other well-known models, such as U-Sleep and YASA.
Researchers compared sleep data with medical records over 25 years, identifying 130 diseases whose risk could be predicted with high accuracy. Rahul Tapa noted that routine sleep measurements open new horizons for analyzing long-term health conditions.
The model predicts dementia, Parkinson's disease, heart attack, heart failure, and certain types of cancer with the highest accuracy. According to Sebastian Buschieger, a sleep expert from the Lamarr Institute, AI can be trained to predict a wide range of diseases if the appropriate database is available.
What AI Looks for in a Sleeping Person's Body
Analysis shows that heart signals play a key role in predicting cardiovascular diseases, while brain signals are crucial for neurological disorders. Interestingly, discrepancies between brain states and heart rhythms may indicate hidden stressors or early stages of diseases. Experts emphasize that the correlations provided by AI are primarily statistical and require validation from medical professionals.
Reliability of Laboratory Data
The model is based on data obtained in sleep laboratories, where patients are often referred due to sleep problems. Researchers from various American and European groups continue to test the model; however, the lack of data from individuals without sleep issues limits its universality.
Potential and Limitations of Diagnosis and Therapy
It is important to note that SleepFM does not identify the causes of diseases, only correlations between various sleep indicators and possible diagnoses. As computer scientist Matthias Jacobs explains, most AI methods are not capable of establishing causal relationships, but even statistical correlations can be useful for diagnosis and therapy.
AI as an Auxiliary Tool for Doctors
Models like SleepFM allow for the rapid processing of large volumes of polysomnography data, facilitating the analysis of sleep stages and the diagnosis of apnea. This approach helps medical professionals focus on patients while leaving routine tasks to AI. However, as Sebastian Buschieger emphasizes, AI remains an auxiliary tool, and final decisions regarding diagnosis and treatment are still made by doctors.
Researchers continue to explore whether the identified patterns can indicate biological mechanisms of diseases. If certain sleep signals are consistently associated with specific diseases, this may point to disruptions in the nervous, cardiovascular, or immune systems in the early stages of illness.
