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

2026-06-13

AI Summary

Existing photoplethysmography (PPG) representation learning approaches are largely confined to the signal level and lack patient-level health context, limiting their generalizability in complex clinical settings and heterogeneous populations. To address this, this work constructs a large-scale multimodal PPG–electronic health record (EHR) dataset and proposes Clinical Anchor Pretraining (CAP), a novel framework that leverages EHR data as patient-level supervisory signals. By employing cross-modal contrastive learning, CAP aligns PPG representations with holistic physiological states captured in EHRs. This approach substantially enhances model interpretability, robustness, and transferability, achieving an average relative performance improvement of 26.7% across four downstream tasks, with respiratory rate prediction showing the highest gain of 87.6%, significantly outperforming current baselines.

Abstract

Photoplethysmography (PPG) plays a central role in wearable health monitoring and clinical decision support. Yet existing approaches to universal PPG representation learning largely focus on signal-level objectives and often overlook patient-level health context, which limits generalization to complex clinical tasks and heterogeneous cohorts. To address this gap, we construct a large-scale paired PPG-EHR multimodal dataset by distilling fragmented medical histories and clinical records into cohesive, patient-level electronic health records (EHR). Building on this resource, we propose Clinical Anchored Pretraining for PPG (CAP). During pretraining, CAP performs cross-modal contrastive alignment that anchors PPG representations to patient-level clinical semantics, guiding the encoder beyond waveform fitting toward modeling consistency in a patient's overall physiological state. During downstream adaptation, the pretrained PPG encoder provides clinically grounded representations that strengthen inductive bias and improve robustness and transferability. Experiments demonstrate that CAP consistently outperforms strong baselines on four diverse downstream tasks. CAP achieves a particularly large gain on respiratory rate prediction (up to +87.6% relative improvement over the state-of-the-art baseline) and delivers an average relative +26.7% across all tasks. We further enhance the interpretability of our approach through comprehensive analyses, including ablations and multiple complementary visualizations of the learned representations. The code for our experiments is available at: https://github.com/gody123gody/CAP .

Problem

• PPG
• universal representation learning
• patient-level supervision
• clinical context
• generalization

Innovation

• PPG representation learning
• patient-level supervision
• multimodal contrastive learning
• clinical semantics alignment
• wearable health monitoring