CAP: Towards PPG Universal Representation Learning with Patient-level Supervision
Researchers have developed a new method called Clinical Anchored Pretraining for PPG (CAP) to improve the learning of universal representations for photoplethysmography (PPG) signals. Existing methods often overlook patient-level health context, limiting their generalization. CAP addresses this by constructing a large-scale paired PPG-EHR multimodal dataset and using cross-modal contrastive alignment to anchor PPG representations to clinical semantics. This approach enhances robustness and transferability, showing significant improvements on downstream tasks, particularly in respiratory rate prediction. AI