Contextualizing remote fall risk: Video data capture and implementing ethical AI.

Area of Science:

• Digital Medicine
• Biomedical Engineering
• Computer Vision

Background:

• Wearable inertial measurement units (IMUs) quantify gait for fall risk but lack contextual data.
• Current IMU gait analysis is limited by the absence of real-world environmental context.
• Ethical and privacy concerns hinder the use of wearable cameras in clinical settings.

Purpose of the Study:

• To propose a privacy-preserving method for integrating wearable cameras with IMUs for fall risk assessment.
• To demonstrate the feasibility of using AI-based computer vision to contextualize IMU gait data.
• To enhance the understanding of habitual fall risk in free-living environments.

Main Methods:

• Utilizing AI-based computer vision models to automatically detect and blur sensitive information (e.g., people) in video data.
• Employing off-the-shelf methods for object detection and blurring to maintain privacy.
• Integrating video analysis with wearable IMU data for comprehensive gait assessment.

Main Results:

• An exemplar AI model achieved 88% accuracy in detecting and blurring sensitive objects.
• The proposed approach enables a more holistic assessment of free-living fall risk.
• Contextual video data can be preserved while effectively obscuring private information.

Conclusions:

• Routine use of wearable cameras for gait analysis is achievable in digital medicine through AI-driven privacy solutions.
• This integrated approach offers a comprehensive understanding of fall risk without compromising patient privacy.
• The AI-powered video and IMU integration has broader applications beyond fall risk assessment for various clinical cohorts.