Validity of a Convolutional Neural Network-Based, Markerless Pose Estimation System Compared to a Marker-Based 3D...

Area of Science:

• Biomechanics
• Computer Vision
• Clinical Motion Analysis

Background:

• Marker-based 3D optical systems are the current standard for gait analysis.
• Markerless systems using Convolutional Neural Networks (CNNs) are emerging as important tools.
• Validation of these new markerless systems against established methods is crucial for clinical adoption.

Purpose of the Study:

• To validate the performance of a CNN-based markerless gait analysis system (Orthelligent VISION; OV).
• To compare OV's accuracy against a standard marker-based 3D motion capture system.
• To assess the clinical applicability of markerless gait analysis.

Main Methods:

• A pilot study involving 16 healthy adults.
• Comparison of the Orthelligent VISION (OV) markerless system with a marker-based 3D motion capture system.
• Analysis of standard gait metrics using concordance correlation coefficients (CCCs) and coefficients of variation.

Main Results:

• Good overlap was found in gait event detection between both systems.
• OV demonstrated strong to almost complete concordance (CCC ≥ 0.624) for sagittal plane metrics (cadence, variability, step/stance time, step length, double support).
• Strong concordance (CCC ≥ 0.805) was observed for frontal plane parameters (cadence, step/stance time, step width), with moderate to weak correlation for gait symmetry.

Conclusions:

• The CNN-based, markerless gait analysis system OV shows high accuracy for many clinical gait metrics.
• OV is a promising alternative to marker-based 3D motion capture systems.
• Markerless gait analysis holds potential for broad clinical applications.