A robust PPG foundation model using multimodal physiological supervision
Researchers have developed a new foundation model for photoplethysmography (PPG) data that enhances robustness by utilizing multimodal physiological signals like electrocardiograms and respiratory data during pretraining. This approach allows the model to learn from noisy PPG segments, improving its generalization to real-world, consumer-grade data without requiring extensive curated datasets. The model achieved performance improvements on 14 out of 15 diverse downstream tasks, including daily activity and heart rate prediction, using significantly fewer subjects than existing state-of-the-art methods. AI
IMPACT Enhances generalization of physiological monitoring models to consumer-grade devices.