A Physics-Data Hybrid Framework Using Uncalibrated Consumer CMOS Vision: Pilot Study on Monocular Automatic TUG...

Area of Science:

• Biomedical Engineering
• Neurology
• Computer Vision

Background:

• The Timed Up and Go (TUG) test is vital for elderly mobility assessment but faces barriers in remote and low-resource settings due to high costs and calibration needs.
• Existing automated mobility assessment tools often require specialized hardware and expert setup, limiting widespread adoption.

Purpose of the Study:

• To develop a "plug-and-play" Physics-Data Hybrid framework for early Parkinson's disease (PD) risk screening using uncalibrated consumer-grade cameras.
• To enable accurate, low-cost, and accessible mobility assessment for telehealth applications, particularly in underserved populations.

Main Methods:

• Integration of learning-based pose perception with a self-evolving physics model to recover metric-scale motion without manual calibration.
• Implementation of a noise-adaptive fusion strategy to combine 2D pixel dynamics with 3D kinematic data, addressing monocular vision's scale ambiguity.
• Extraction of high-dimensional spatiotemporal gait parameters (e.g., stride length variation, gait velocity) for enhanced diagnostic resolution.

Main Results:

• The framework achieved 98% screening accuracy and 87.32% overall classification accuracy in a pilot study of 10 subjects.
• Extracted metric gait features effectively identified risk staging in simulated dual-task cognitive-motor interference scenarios.
• Demonstrated the ability to capture subtle motor fluctuations indicative of PD risk.

Conclusions:

• The Physics-Data Hybrid framework offers a robust, low-cost solution for early PD risk assessment via telehealth.
• This technology can significantly improve accessibility to neurological screening for underserved populations.
• The system overcomes limitations of traditional TUG tests, providing finer diagnostic detail for early motor impairment detection.