Accuracy of Monocular Two-Dimensional Pose Estimation Compared With a Reference Standard for Kinematic Multiview...

Area of Science:

• Biomedical Engineering
• Clinical Biomechanics
• Medical Technology

Background:

• Expensive optoelectronic systems are the gold standard for kinematic analysis but are impractical for routine clinical use.
• Smartphone-based 2D human pose estimation offers a more accessible approach for clinical kinematic analysis.
• Evaluating the accuracy of these apps is crucial for clinical specialists who rely on valid data.

Purpose of the Study:

• To assess the accuracy of the Lindera-v2 mobility analysis app for 2D human pose estimation.
• To compare joint angle measurements from the Lindera-v2 app against a validated reference standard (PanopticStudio Toolbox).
• To determine the clinical applicability of the Lindera-v2 app based on its accuracy.

Main Methods:

• Analyzed 10 video sequences using the Lindera-v2 algorithm.
• Compared 30,000 data pairs per joint (10 joints) between the app and the reference standard.
• Calculated mean differences, mean absolute error (MAE), and intraclass correlation coefficients (ICC) to quantify accuracy and agreement.
• Performed cross-correlation analysis to detect temporal lags.

Main Results:

• The Lindera-v2 app demonstrated excellent agreement with the reference standard, with ICC values ranging from 0.951 to 0.997.
• Mean angle differences were minimal, with the right hip showing the closest agreement (-0.05°).
• The algorithm showed no temporal lag, indicating accurate time-series data.

Conclusions:

• Smartphone-based 2D pose estimation, as exemplified by the Lindera-v2 app, achieves excellent agreement with validated reference standards.
• The Lindera-v2 algorithm provides accurate kinematic variable assessment with minimal deviations compared to complex systems.
• This technology presents a promising, accessible tool for clinical mobility analysis.