Continuous Cardiac Arrest Prediction in ICU using PPG Foundation Model

2025-02-12

AI Summary

This study addresses the clinical need for early, non-invasive prediction of in-hospital cardiac arrest (IHCA) in intensive care units (ICUs). We propose a continuous, single-channel photoplethysmography (PPG)-based forecasting framework that operates exclusively on fingertip PPG signals. Our method introduces a two-stage Feature Extraction–Aggregation Network (FEAN), which—novelty—integrates a billion-parameter PPG-GPT foundation model with a lightweight temporal classifier to enable deep waveform representation learning and dynamic risk modeling. We design dual sliding-window schemes—“1H” (1-hour) and “FH” (full-horizon, up to 24 hours)—to support multi-granularity early warning. Evaluated on real-world ICU data, our model achieves an average 24-hour AUROC of 0.79 and a peak 1-hour-ahead AUROC of 0.82. PaCMAP-based latent-space visualization further enables interpretable patient health trajectory analysis. To our knowledge, this is the first work achieving high-accuracy, long-horizon, and interpretable IHCA prediction using only continuous, unimodal PPG signals.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Activity and Plan Recognition
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• 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
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Non-invasive patient monitoring for tracking and predicting adverse acute health events is an emerging area of research. We pursue in-hospital cardiac arrest (IHCA) prediction using only single-channel finger photoplethysmography (PPG) signals. Our proposed two-stage model Feature Extractor-Aggregator Network (FEAN) leverages powerful representations from pre-trained PPG foundation models (PPG-GPT of size up to 1 Billion) stacked with sequential classification models. We propose two FEAN variants ("1H","FH") which use the latest one-hour and (max) 24-hour history to make decisions respectively. Our study is the first to present IHCA prediction results in ICU patients using only unimodal (continuous PPG signal) waveform deep representations. With our best model, we obtain an average of 0.79 AUROC over 24~h prediction window before CA event onset with our model peaking performance at 0.82 one hour before CA. We also provide a comprehensive analysis of our model through architectural tuning and PaCMAP visualization of patient health trajectory in latent space.

Problem

• Predict in-hospital cardiac arrest using PPG signals
• Leverage pre-trained PPG foundation models for accuracy
• Analyze patient health trajectory with deep representations

Innovation

• PPG foundation model
• Feature Extractor-Aggregator Network
• Unimodal waveform deep representations