Assessing Time-Varying Lumbar Flexion-Extension Kinematics Using Automated Pose Estimation

Area of Science:

• Biomechanics
• Computer Vision
• Kinesiology

Background:

• 3D motion capture systems like Vicon are gold standards for kinematic analysis.
• DeepLabCut (DLC) is an emerging markerless pose estimation tool with potential for biomechanical applications.
• Validation of DLC against established systems is crucial for its adoption in research.

Purpose of the Study:

• To compare the accuracy and precision of DeepLabCut (DLC) with a Vicon 3D motion capture system.
• To evaluate DLC's performance in analyzing lumbar and elbow flexion-extension movements.
• To assess DLC's reliability across different datasets (training and testing groups).

Main Methods:

• Concurrent data acquisition using DLC and Vicon systems.
• Training a novel DLC model on video data from a subset of participants.
• Assessing accuracy and precision using both training and independent testing datasets.
• Statistical analysis (two-way ANOVA) and Bland-Altman plots for comparison.

Main Results:

• No significant differences were found between DLC and Vicon in estimating planar angles for lumbar and elbow movements.
• No significant interactions were observed between motion capture modality and dataset (training vs. testing).
• Bland-Altman plots indicated acceptable agreement and low bias between DLC and Vicon.

Conclusions:

• DeepLabCut provides accurate and precise planar kinematic data for the elbow and lumbar spine.
• DLC is a viable, validated alternative to traditional 3D motion capture for specific biomechanical analyses.
• The findings support the use of DLC in research requiring accessible and reliable kinematic measurements.