PaPaGei: Open Foundation Models for Optical Physiological Signals

2024-10-27

arXiv.org

AI Summary

Existing PPG-based machine learning models exhibit strong task specificity and poor generalizability, hindered by reliance on single-device data, absence of out-of-domain validation, and lack of open-source models—resulting in low reproducibility. Method: We propose the first open-source foundation model for PPG signals, introducing a novel self-supervised representation learning paradigm that integrates morphological priors. Our approach leverages 20 million unlabeled PPG segments (57,000 hours), employing physiology-guided contrastive augmentation and signal reconstruction as joint pretraining objectives, within a lightweight, efficient architecture. Contribution/Results: The model achieves cross-device, cross-population, and cross-health-scenario generalization and establishes a skin-tone-robustness evaluation benchmark. On 10 datasets across 20 tasks, it improves classification and regression performance by 6.3% and 2.9%, respectively, significantly outperforming prior methods on 14+ tasks. With only 1/70 the parameters of state-of-the-art large models, it supports both feature extraction and multimodal fusion.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Photoplethysmography (PPG) is the leading non-invasive technique for monitoring biosignals and cardiovascular health, with widespread adoption in both clinical settings and consumer wearable devices. While machine learning models trained on PPG signals have shown promise, they tend to be task-specific and struggle with generalization. Current research is limited by the use of single-device datasets, insufficient exploration of out-of-domain generalization, and a lack of publicly available models, which hampers reproducibility. To address these limitations, we present PaPaGei, the first open foundation model for PPG signals. The model is pre-trained on over 57,000 hours of data, comprising 20 million unlabeled PPG segments from publicly available datasets. We introduce a novel representation learning approach that leverages domain knowledge of PPG signal morphology across individuals, enabling the capture of richer representations compared to traditional contrastive learning methods. We evaluate PaPaGei against state-of-the-art time-series foundation models and self-supervised learning benchmarks across 20 tasks from 10 diverse datasets, spanning cardiovascular health, sleep disorders, pregnancy monitoring, and wellbeing assessment. Our model demonstrates superior performance, improving classification and regression metrics by 6.3% and 2.9% respectively in at least 14 tasks. Notably, PaPaGei achieves these results while being more data- and parameter-efficient, outperforming models that are 70x larger. Beyond accuracy, we examine model robustness across different skin tones, establishing a benchmark for bias evaluation in future models. PaPaGei can serve as both a feature extractor and an encoder for multimodal models, opening up new opportunities for multimodal health monitoring.

Problem

• Generalizing machine learning models for PPG signals
• Addressing limitations in PPG dataset diversity
• Developing open, efficient foundation models for PPG

Innovation

• Open foundation model for PPG signals
• Novel representation learning approach
• Data- and parameter-efficient design