Inter-system differences in modelling approaches for shoulder joint kinematics between markerless and marker-based...

Comparative data on markerless motion capture systems for upper-extremity kinematics during high-speed movements remain limited. We aimed to examine inter-system modelling differences in shoulder joint kinematics between a markerless motion capture system (Theia3D) and a marker-based system during baseball pitching, focusing on a wide range of shoulder joint motion throughout baseball pitching. Six collegiate pitchers completed 117 trials recorded simultaneously by both systems. Shoulder joint angles and joint centre positions were compared using nested Bland-Altman analysis, intraclass correlation coefficients (ICC (3,1)), and statistical parametric mapping (SPM). At maximum external rotation (MER), inter-system differences were small for external rotation (bias: -0.3°; ICC = 0.77), whereas the horizontal angle demonstrated a pronounced fixed bias (bias: -19.1°). At ball release (BR), inter-system differences increased across all joint angles, with the largest discrepancy for external rotation (bias: -8.5°; ICC = 0.33). Proportional bias was evident for most variables at BR. Between-participant variance exceeded within-participant variance at MER, indicating that participant-specific offsets were primary contributors to the limits of agreement. SPM identified significant inter-system differences during the acceleration phase. These findings highlight the importance of understanding the characteristics of each modelling approach for appropriate interpretation and application of markerless motion capture in overhead athletes.