Options
March 2025
Journal Article
Title
Comparison of markerless and marker-based 3D gait analysis of gait perturbations on a perturbation treadmill
Abstract
Background:
Markerless 3D gait analysis (3DGA) is a method for quantitative motion analysis that is based on AI-supported learning techniques. Artificial neural networks are employed to predict the position of key body landmarks on image or video data. Compared to marker-based 3DGA, the markerless method offers the advantage of enhanced flexibility and user-friendliness. Measuring spatiotemporal gait parameters with markerless 3DGA has already shown promising results for individuals with cerebral palsy, post-stroke patients, and healthy subjects [1] [2] [3]. Data from markerless 3DGA for gait perturbations on a perturbation-treadmill are not yet available. Therefore, this study aimed to compare the validity and reliability of markerless 3DGA with a marker-based system for gait perturbations on a perturbation-treadmill in healthy young adults.
Methods:
Consecutive consenting healthy adults (age 18-55; n=50) were recruited for this monocentric cross-sectional study. All participants walked on a dual-belt treadmill (M-Gait, Motek Medical BV, NL), while being exposed a total of 8 perturbations, consisting of slips and trips induced by unilateral belt decelerations during heel strike (HS) and mid-stance, as well as sways during the single limb support phase. Before data collection, participants were familiarized with walking on the treadmill in a self-selected speed for approximately 6 minutes. For reliability measurements, participants repeated the task after 1-7 days. All sessions were simultaneously recorded by a synchronized markerless and marker-based system. Data processing was performed using Vicon Nexus (Vicon Motion Systems Ltd., UK) for the marker-based data and Theia3D (Theia Markerless Inc., CA) for the markerless data. The primary outcomes were kinematic variables and gait parameters derived from the marker trajectories and RGB data using inverse kinematics.
Results:
To date 11 healthy adults (7 women, aged 31,7±7,0 years) completed a total of 20 measurements. The mean (±SD) deviation of the of the hip, knee and ankle coordinates |Δφx| in the sagittal plane between the markerless and marker-based system is plotted in Figure 1. For normal steps, the mean (±SD) deviation is presented over the entire gait cycle. The mean (±SD) deviation during left HS perturbation was analysed from the perturbation starting point tlhs0 to the endpoint tlhs1. The mean deviation of the hip, knee and ankle coordinates in the sagittal plane for normal left steps and left HS perturbation ranged between 2 and 12 degrees between both conditions.
Markerless 3D gait analysis (3DGA) is a method for quantitative motion analysis that is based on AI-supported learning techniques. Artificial neural networks are employed to predict the position of key body landmarks on image or video data. Compared to marker-based 3DGA, the markerless method offers the advantage of enhanced flexibility and user-friendliness. Measuring spatiotemporal gait parameters with markerless 3DGA has already shown promising results for individuals with cerebral palsy, post-stroke patients, and healthy subjects [1] [2] [3]. Data from markerless 3DGA for gait perturbations on a perturbation-treadmill are not yet available. Therefore, this study aimed to compare the validity and reliability of markerless 3DGA with a marker-based system for gait perturbations on a perturbation-treadmill in healthy young adults.
Methods:
Consecutive consenting healthy adults (age 18-55; n=50) were recruited for this monocentric cross-sectional study. All participants walked on a dual-belt treadmill (M-Gait, Motek Medical BV, NL), while being exposed a total of 8 perturbations, consisting of slips and trips induced by unilateral belt decelerations during heel strike (HS) and mid-stance, as well as sways during the single limb support phase. Before data collection, participants were familiarized with walking on the treadmill in a self-selected speed for approximately 6 minutes. For reliability measurements, participants repeated the task after 1-7 days. All sessions were simultaneously recorded by a synchronized markerless and marker-based system. Data processing was performed using Vicon Nexus (Vicon Motion Systems Ltd., UK) for the marker-based data and Theia3D (Theia Markerless Inc., CA) for the markerless data. The primary outcomes were kinematic variables and gait parameters derived from the marker trajectories and RGB data using inverse kinematics.
Results:
To date 11 healthy adults (7 women, aged 31,7±7,0 years) completed a total of 20 measurements. The mean (±SD) deviation of the of the hip, knee and ankle coordinates |Δφx| in the sagittal plane between the markerless and marker-based system is plotted in Figure 1. For normal steps, the mean (±SD) deviation is presented over the entire gait cycle. The mean (±SD) deviation during left HS perturbation was analysed from the perturbation starting point tlhs0 to the endpoint tlhs1. The mean deviation of the hip, knee and ankle coordinates in the sagittal plane for normal left steps and left HS perturbation ranged between 2 and 12 degrees between both conditions.
Author(s)