BCG-FM: A Foundation Model for Ambient Cardiac Health Sensing
Researchers have developed BCG-FM, a novel foundation model for analyzing cardiac health through ambient mechanical biosignals. This model utilizes a piezoelectric sensor embedded in a bed surface to record ballistocardiography (BCG) data overnight, requiring no user effort. Pretrained on 2.75 million hours of recordings from nearly 146,000 individuals, BCG-FM achieved a 3.26-year Mean Absolute Error in biological age estimation and demonstrated clinically relevant discrimination across various health conditions. AI
IMPACT Introduces a new, passive data modality for foundation models in healthcare, potentially enabling continuous, effortless health monitoring.