Comparing Theia3D software versions for gait kinematics in typically developing children

Markerless motion capture systems such as Theia3D offer an alternative to traditional marker-based approaches for measuring gait kinematics. As the underlying algorithms continue to evolve, successive software updates may introduce differences in kinematic outputs. However, the extent of these differences is not well known. The purpose of this study was to quantify inter-version differences in Theia3D-derived gait kinematics in typically developing children. Sixteen typically developing children completed five barefoot walking trials at a self-selected speed. Video data were processed using three versions of Theia3D (v2022.1.0.2309, v2023.1.03161 and v2024.1.24). Inter-version differences were assessed across joints and planes of motion using one-dimensional Statistical Parametric Mapping (1D-SPM) repeated measures ANOVA, with planned pairwise comparisons between sequential releases. Significant differences were identified across all joints and planes of motion. The largest differences were observed between v2022 and v2023, with smaller and more localised differences between v2023 and v2024. Mean differences were generally below 6°, with larger deviations observed in the transverse plane at the ankle and foot and in the sagittal plane at the trunk, pelvis and hip. Differences were not consistent across versions, with some angles demonstrating patterns that reverted towards earlier outputs. These findings demonstrate that software version does influence motion capture kinematics. While differences are small at some joint levels, there is potential for software version to influence interpretation in certain contexts, particularly in longitudinal assessments. Obligatory reporting of software version and data reprocessing are recommended when using markerless systems.