AI Predicts Disease from One Night of Sleep (www.sciencedaily.com)

Stanford researchers have introduced SleepFM, an innovative AI system capable of predicting disease risk from just a single night of sleep data. By analyzing nearly 600,000 hours of polysomnography recordings, the AI identifies hidden patterns within various physiological signals—such as brain activity, heart function, and breathing—revealing insights that suggest sleep itself may be a crucial, yet neglected, source of early health warnings. With this technology, the AI can forecast the risk of over 130 medical conditions, including cancer and dementia, offering significant implications for preventive healthcare. This study is groundbreaking as it leverages artificial intelligence to harness the vast amounts of data captured during sleep, a domain previously underexplored in AI research. The model is designed to learn the "language of sleep" through advanced training techniques, enabling it to effectively combine diverse physiological signals to enhance predictive accuracy. Notably, SleepFM achieved high performance metrics, with a concordance index (C-index) of 0.89 for Parkinson's disease and 0.87 for breast cancer, indicating strong potential for future applications in personal health monitoring and early disease detection. The researchers aim to further improve the model's predictions and understanding, potentially integrating wearable device data for even broader healthcare implications.