Validity and Reliability of OpenPose-Based Motion Analysis in Measuring Knee Valgus during Drop Vertical Jump Test

Area of Science:

• Biomechanics
• Computer Vision
• Deep Learning

Background:

• Conventional 3D motion analysis (3D-MA) and human visual detection-based motion analysis (Human-MA) for human motion estimation are limited by cost, time, and experimental constraints.
• OpenPose-based motion analysis (OpenPose-MA) leverages deep learning for 2D pose estimation, offering a potentially more accessible approach.

Purpose of the Study:

• To evaluate the precision of OpenPose-MA against Human-MA, using 3D-MA as the gold standard.
• To compare the reproducibility, accuracy, and waveform similarity of knee valgus angle measurements between OpenPose-MA and Human-MA.

Main Methods:

• A cohort of 21 healthy adults performed a drop vertical jump task.
• Knee valgus angles were measured using OpenPose-MA (2D pose estimation) and Human-MA (physiotherapist assessment).
• Three-dimensional motion analysis (3D-MA) served as the reference standard for comparison.

Main Results:

• OpenPose-MA and Human-MA showed excellent reproducibility (ICCs) comparable to 3D-MA.
• No significant differences were found between OpenPose-MA and Human-MA in mean absolute error (MAE) or coefficient of multiple correlation (CMC) for knee valgus angles.
• Pearson's correlation coefficients indicated strong agreement between OpenPose-MA (0.97) and Human-MA (0.98) with 3D-MA.

Conclusions:

• OpenPose-MA provides satisfactory reproducibility and accuracy for knee valgus angle assessment, comparable to Human-MA.
• Deep learning-based OpenPose-MA demonstrates waveform similarity to 3D-MA, similar to Human-MA.
• OpenPose-MA is a viable, accurate, and reproducible method for analyzing knee valgus during dynamic movements like the drop vertical jump.