An AI can predict the risk of developing more than 100 diseases with just one night of sleep.

A model developed by Stanford University uses advanced polysomnography data to identify physiological patterns and predict the future risk of conditions such as dementia, heart attack, and atrial fibrillation.

                                                                                             

The study published in Nature Medicine highlights that artificial intelligence can interpret large volumes of sleep data that are impossible for humans to analyze.

      

 An artificial intelligence (AI) system developed by Stanford University is capable of predicting the risk of more than 100 diseases based on data collected during a single night of sleep, which could transform the early detection of serious illnesses in the future.

The model, called SleepFM, uses information obtained through polysomnography, considered the most advanced standard in sleep analysis, to build a profile that allows it to anticipate the possible onset of conditions such as dementia, myocardial infarction, heart failure, chronic kidney disease, stroke, and atrial fibrillation, according to the journal Nature Medicine.

The Californian university, one of the most renowned in the world, is thus advancing a line of research aimed at unraveling the impact of artificial intelligence on preventive medical diagnosis.

How this artificial intelligence model works in the healthcare sector.

                                                                                            

SleepFM has been trained with 600,000 hours of sleep data from 65,000 participants, processing brain, heart, muscle, and respiratory signals.

The development of SleepFM represents the first scientific effort to use artificial intelligence to analyze sleep data on a large scale. The tool has been trained with nearly 600,000 hours of data collected from 65,000 participants, processing brain, heart, muscle, and respiratory signals.

The study text emphasizes: “SleepFM produces latent representations of sleep that capture the physiological and temporal structure of sleep and enable accurate prediction of the risk of future diseases.”

The researchers highlight that advances in AI make it possible to address the challenge of interpreting the vast amounts of data generated by polysomnography, which until now has been too complex for traditional human analysis.

                                                                                 

The SleepFM model uses polysomnography data to predict serious conditions such as dementia, myocardial infarction, and chronic kidney disease.

For his part, James Zou, associate professor of biomedical data science and co-author of the study, emphasized in a statement to the Stanford School of Medicine website: "From an AI perspective, sleep is relatively understudied."

What other AI models have been created to support human doctors

Internationally, other groups have presented complementary advances. A group of scientists has developed the Delphi-2M model, a tool that, using natural language processing technologies similar to ChatGPT, is capable of analyzing medical records and predicting the onset of more than 1,000 future diseases.

                                                                                            

The Delphi-2M model, based on artificial intelligence, analyzes medical records and predicts more than 1,000 future diseases.

Delphi-2M's operation is based on the study of sequences of medical diagnoses, allowing it to identify "patterns in health data, previous diagnoses, the combinations in which they occur, and their sequence," explained Moritz Gerstung, an artificial intelligence expert at the German Cancer Research Center.

To train Delphi-2M, researchers used the UK Biobank biomedical database, which contains information from 500,000 participants in Great Britain, and validated its performance with public medical records corresponding to nearly two million people in Denmark.

What is the true potential of AI in the healthcare sector

During a recent competition, four heads of department from local hospitals competed anonymously with two AI platforms, the Chinese Gastrointestinal Multimodal AI and the Claude model, to solve a complex digestive clinical case by analyzing endoscopic images, additional tests, and CT scans.

                                                                                       

In a competition, AI systems achieved diagnoses in seconds, matching the accuracy of doctors in complex clinical cases.

While the doctors required approximately 13 minutes to reach a consensus on a diagnosis and define a treatment plan, the artificial intelligence systems offered a proposal in less than two seconds.

The platform developed by Shanghai AI Lab, trained with 30,000 real clinical cases, completely matched the assessment of the human professionals, but the Claude model, according to the organizers, showed minor differences in the interpretation of the tests, which were reflected in its success rate.

These findings highlight the ability of artificial intelligence to process large volumes of clinical data at a speed impossible for human medical specialists.