AURA: Development and Validation of an Augmented Unplanned Removal Alert System using Synthetic ICU Videos

2025-11-15

AI Summary

This study addresses unplanned extubation (UE)—a critical patient safety issue in intensive care units (ICUs)—by proposing a privacy-compliant, real-time visual alerting method. Methodologically, it introduces the first application of text-to-video diffusion models to generate high-fidelity, diverse synthetic ICU video data, thereby eliminating privacy and ethical concerns associated with real patient imagery. A pose estimation model trained exclusively on this synthetic data enables precise anatomical keypoint tracking, facilitating detection of hand-to-airway proximity events and quantitative assessment of patient agitation. Experimental results demonstrate high accuracy in collision detection and moderate performance in agitation classification. Clinical expert evaluation confirms the high clinical realism of the synthetic videos. This work establishes a novel, reproducible, and scalable paradigm for medical AI that requires zero real-world patient data, offering strong potential for clinical deployment.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Humans and AI: AI for Accessibility
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization
• Security and Privacy: Data transparency and provenance
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Unplanned extubation (UE) remains a critical patient safety concern in intensive care units (ICUs), often leading to severe complications or death. Real-time UE detection has been limited, largely due to the ethical and privacy challenges of obtaining annotated ICU video data. We propose Augmented Unplanned Removal Alert (AURA), a vision-based risk detection system developed and validated entirely on a fully synthetic video dataset. By leveraging text-to-video diffusion, we generated diverse and clinically realistic ICU scenarios capturing a range of patient behaviors and care contexts. The system applies pose estimation to identify two high-risk movement patterns: collision, defined as hand entry into spatial zones near airway tubes, and agitation, quantified by the velocity of tracked anatomical keypoints. Expert assessments confirmed the realism of the synthetic data, and performance evaluations showed high accuracy for collision detection and moderate performance for agitation recognition. This work demonstrates a novel pathway for developing privacy-preserving, reproducible patient safety monitoring systems with potential for deployment in intensive care settings.

Problem

• Detects unplanned extubation risks in ICUs using synthetic video data
• Identifies high-risk patient movements like collision and agitation patterns
• Addresses privacy concerns by eliminating need for real patient videos

Innovation

• Uses synthetic ICU videos for training
• Leverages text-to-video diffusion generation
• Applies pose estimation for movement analysis