Comparison of 2D-pose with 3D motion capture methods for assessment of active shoulder range of motion

Smartphone-based 2D-pose estimation offers a convenient method for assessing shoulder range-of-motion (ROM), but its accuracy compared to 3D motion capture (3D-mocap) needs to be determined.

Shoulder ROM was recorded in seventeen participants without shoulder issues using 3D-mocap and 2D-pose concurrently. Movements included abduction, flexion, extension, external, and functional internal rotation (IR). 2D-pose ROM estimates (mymobility's^® Skeletal Tracking Shoulder Range of Motion Assessments feature (Apple Vision framework, Apple Inc., Cupertino, CA, USA) were compared to 3D-mocap using linear mixed-models and Bland-Altman analysis. The influence of thoracic compensation and anatomical frame definitions on shoulder ROM estimates was examined.

High consistency was observed between 2D-pose and 3D-mocap (R ² > 0.98), especially for abduction and flexion. Differences in ROM were linked to anatomical frame variations and thoracic contributions, with 2D-pose overestimating ROM at greater ranges (2°-25°). Internal rotation zone identification showed high consistency, but 2D-pose-based extension and external rotation showed more variability due to thoracic compensation.

Smartphone-based 2D-pose estimation provides a valid alternative for shoulder ROM measurement but should not be used interchangeably with 3D-mocap due to discrepancies arising from anatomical frame definitions and thoracic movements. Shoulder ROM assessment requires consideration of these limitations to ensure appropriate clinical interpretation.