CAP: Towards PPG Universal Representation Learning with Patient-level Supervision

The article presents the Clinical Anchored Pretraining for PPG (CAP), a novel approach for universal representation learning from photoplethysmography (PPG) data that incorporates patient-level supervision. CAP utilizes a large-scale paired PPG-electronic health records (EHR) dataset to perform cross-modal contrastive alignment, significantly improving the model's ability to generalize across clinical tasks. Benchmarks indicate that CAP achieves up to a 87.6% relative improvement in respiratory rate prediction and an average of 26.7% across various tasks, enhancing robustness and transferability for practitioners working with PPG data in healthcare applications.