GPT-PPG: A GPT-based Foundation Model for Photoplethysmography Signals

2025-03-11

AI Summary

This study addresses the absence of general-purpose foundation models for photoplethysmography (PPG) signals. We propose PPG-GPT—the first large-scale, PPG-specific generative model—adapting the GPT architecture to continuous physiological time-series modeling. Pretrained on 210 million 30-second PPG segments, PPG-GPT introduces three key innovations: optimized positional encoding for physiological dynamics, continuous-value embedding for raw signal representation, and hierarchical temporal attention to capture multi-scale physiological patterns. The model exhibits dual discriminative (e.g., atrial fibrillation detection) and generative (e.g., end-to-end signal denoising) capabilities: it achieves high-fidelity denoising without fine-tuning and matches or surpasses state-of-the-art performance on supervised tasks such as AF detection after task-specific fine-tuning. Extensive experiments demonstrate strong cross-dataset generalization and cross-task transferability, establishing a new paradigm for wearable physiological signal analysis.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Diffusion Models for Vision

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
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

This study introduces a novel application of a Generative Pre-trained Transformer (GPT) model tailored for photoplethysmography (PPG) signals, serving as a foundation model for various downstream tasks. Adapting the standard GPT architecture to suit the continuous characteristics of PPG signals, our approach demonstrates promising results. Our models are pre-trained on our extensive dataset that contains more than 200 million 30s PPG samples. We explored different supervised fine-tuning techniques to adapt our model to downstream tasks, resulting in performance comparable to or surpassing current state-of-the-art (SOTA) methods in tasks like atrial fibrillation detection. A standout feature of our GPT model is its inherent capability to perform generative tasks such as signal denoising effectively, without the need for further fine-tuning. This success is attributed to the generative nature of the GPT framework.

Problem

• Adapt GPT for photoplethysmography signal analysis.
• Pre-train model on 200M PPG samples.
• Achieve SOTA in atrial fibrillation detection.

Innovation

• GPT model adapted for PPG signal analysis
• Pre-trained on 200M+ 30s PPG samples
• Generative tasks like denoising without fine-tuning