Key points are not available for this paper at this time.
Marker-based motion capture is the clinical standard for gait analysis, requiring precise marker placement on anatomical landmarks. This process is time-consuming and prone to human error. Theia3D, a markerless system using machine learning and neural networks, tracks features from 2D video to produce 3D motion analysis, but has yet to be clinically validated, and its use for children is minimal. This study compared markerless system (Theia3D) joint tracking with currently the most widely-used marker-based model in clinical gait analysis (Conventional Gait Model, CGM1.1) in typically developing children and adults. Twenty-three children and 34 adults underwent gait assessments at Alder Hey Children's Hospital, where data from both systems were collected synchronously. Kinematics, kinetics and segment lengths were calculated from both systems. Model differences were quantified using pairwise root mean square deviations (RMSD) during phases that were statistically significantly different as determined by statistical parametric mapping. Segment length differences produced by each model were assessed by mean difference, standard error of the mean and minimal detectable change. Significant differences were observed across the gait cycle in all but one joint levels and planes, with RMSDs up to 8.5° in the sagittal plane, 5.3° in the frontal plane and 10.2° in the transverse plane. Theia3D produced larger peak knee moments in the sagittal and frontal plane compared to the CGM1.1 model and produced shorter segment lengths. This study shows the potential of the developing Theia3D's software in clinical gait analysis with children and adults but emphasises the need for further investigations across populations.
Kearney et al. (Fri,) studied this question.