Assessment of a novel deep learning-based marker-less motion capture system for gait study

Area of Science:

• Biomechanics
• Medical Technology
• Computer Science

Background:

• Marker-less gait analysis systems using deep learning offer significant potential for clinical applications.
• A recently developed marker-less system shows promise for joint center estimation but requires evaluation for gait outcomes.

Purpose of the Study:

• To compare a novel marker-less gait analysis system against a marker-based reference system.
• To assess the accuracy of the marker-less system in measuring clinically relevant gait parameters.

Main Methods:

• A deep learning model was trained on data from 41 subjects performing 10 walking trials each.
• The marker-less system estimated 3D joint center positions and was evaluated against a marker-based system.
• Comparisons included joint position differences, gait event detection, spatiotemporal and kinematic parameters, and inter-trial reliability.

Main Results:

• The marker-less system estimated joint centers with a mean difference of 13.1 mm.
• Gait events were detected with 99% accuracy within 10 ms of reference values.
• Spatiotemporal parameters showed zero bias, and kinematic parameters demonstrated comparable reliability to marker-based systems.

Conclusions:

• The marker-less system can measure spatiotemporal gait parameters within the detectable range of marker-based systems.
• The system shows promising results for several kinematic parameters, with potential for clinical gait analysis.