AURA: Augmented Unplanned Removal Alert

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: (a) collision, defined as hand entry into spatial zones near airway tubes, and (b) 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.

Why is AURA important?

Unplanned extubation (UE) affects 6.7% of ventilated ICU patients, leading to severe complications
Nurse unavailability has been associated with UE incidents, but 24/7 monitoring is not feasible
Video monitoring depends on human attention and operates passively
Privacy concerns limit development of vision-based detection systems

Privacy-Preserving

No real patient footage is used at the pre-deployment stage. AURA supports ethical model development and validation for the privacy sensitive ICU settings. AURA demonstrates a novel pathway for developing vision-based patient safety monitoring system.

Reproducible & Scalable

The AURA dataset is open-source and publicly available for reproduction and further research on Zenodo. The source code is accessible on GitHub. The system utilizes anatomical keypoints and can be adapted for other related applications (e.g., central line).

Clinical Relevance

AURA was assessed by 9 ICU nurses to ensure the clinical relevance of the videos and feasibility of the system. Unlike historical data-driven prediction models and wearable sensors, AURA serves as a real-time, unobtrusive safety monitoring system.

How was AURA developed?

Figure 1: Overview or illustration from the paper

Synthetic Video Generation

AURA dataset comprises 75 synthetic ICU videos made with text-to-video AI model. An experienced ICU nurse carefully designed the prompts to create realistic and clinically relevant scenarios.

System Development

AURA is developed to detect two high-risk movement patterns (Collision and Agitation) with pose estimation model. The expert tunned the parameters with tuning set and applied to the unseen test set.

System Assessment

AURA was assessed by 9 ICU nurses. Video quality, alarm appropriateness, system feasibility are assessed in the expert evaluation. Subsequent performance evaluation is conducted by ground truth annotation.

Result

Expert assessments confirmed the realism of synthetic data (All domains ≥ 4.3). AURA achieved near-perfect collision detection (F1=0.98) and moderate agitation detection (F1=0.78).

See more details in our paper

Citation

@article{seo2025aura,
      title={AURA: Development and Validation of an Augmented Unplanned Removal Alert System using Synthetic ICU Videos},
      author={Seo, Junhyuk and Moon, Hyeyoon and Jung, Kyu-Hwan and Oh, Namkee and Kim, Taerim},
      journal={arXiv preprint arXiv:2511.12241},
      year={2025}
    }