GlucoFM: Foundation model for continuous glucose monitoring

GlucoFM is a foundation model, a type of artificial intelligence system, designed to analyze continuous glucose monitoring data from sensors worn under the skin. The model uses a dual-stream approach that separately tracks slower glucose trends and short-term changes, which allows it to better understand glucose patterns than previous systems that treated all glucose data as a single stream. This matters because understanding glucose dynamics can help assess diabetes risk, insulin resistance, and other metabolic conditions, but high-quality labeled data needed to train such systems is sparse and costly to obtain. The model was pre-trained on over 100,000 hours of unlabeled glucose data and showed stronger performance than existing approaches across multiple prediction tasks and different patient cohorts.