Concurrent validity of Theia3D markerless motion capture compared to a marker-based system for three-dimensional gait analysis in individuals with Parkinson’s disease

Aim and Research Question(s)

• Aim: Concurrent validity of Theia3D (markerless) vs. Vicon (marker-based) for 3D gait analysis in Parkinson's disease (PD)
• RQ1: How well do trunk & lower-limb joint-angle trajectories and spatiotemporal parameters from Theia3D agree with Vicon in PD?
• RQ2: Is agreement affected by the more- vs. less-affected side?

Background

• Marker-based motion capture = gold standard, but resource- & time-intensive; markerless systems (Theia3D) estimate kinematics from video → more accessible
• Good–excellent agreement in healthy adults [1]; 3D validation in PD still lacking [2]

Methods

• Secondary data from EyeQTrack project (USTP, 2023); n = 14 (Hoehn & Yahr I–III; age 68.6±8.8y; UPDRS-III 19.8±10.4)
• Overground walking at self-selected speed; simultaneous recording
• Vicon (22 IR cameras, 55-marker CGM2.5) vs. Theia3D (8 RGB cameras, ~550 gait cycles)
• 36 joint-plane combinations + 30 spatiotemporal parameters; RMSE, Pearson's r, ICC(2,1), Bland-Altman, SPM1D (threshold RMSE > 5°)

Results and Discussion

• Spatiotemporal parameters: excellent validity across all 30 (ICC 0.945-1.000)
• Kinematics strongly plane-dependent (see matrix): sagittal knee, hip & foot progression excellent (RMSE < 5°); frontal & transverse predominantly poor
• SPM1D: significant differences in 26 of 36 combinations
• Pelvis sagittal = model-level incompatibility (CGM2 vs. Theia3D), not a tracking error [1]
• No consistent effect of motor asymmetry

Conclusion

• Theia3D = valid, practical alternative for spatiotemporal parameters & sagittal lower-limb kinematics (knee, hip, foot progression)
• Interpret with caution/not interchangeable: frontal & transverse kinematics, ankle sagittal, pelvis & thorax
• Findings preliminary → replication in larger PD cohorts warranted

References

[1] Kanko et al. (2021). Journal of Biomechanics, 127, 110665. [2] Seuthe et al. (2025). Journal of Biomechanics, 193, 113008.