Anatomy of a pose: Relationships between human shoulder bone shape and pose

Shoulder anatomy is complex, varying in shape and pose. Studies have related bone shape and joint function, which can predict each other but not yet in the shoulder. This project aims to investigate bone shape and pose relationships in the healthy shoulder via partial least square regressions (PLSR). Sixty-eight registered humeri, scapulae, and clavicles were segmented (47 males and 15 females, age: 30.7 ± 9.5 years) from medical images (computed tomography and magnetic resonance images) and constituted the shape input (principal component analysis scores). Local bone coordinate systems (three axes and origin) composed the pose information. PLSR analyses were conducted using the shape of one or all three bones to predict the pose and vice versa. The main variation mode explained scaling for shape (38% variation explained) and the three shoulder bones' anteroposterior (AP) and superoinferior translations, and humeral and clavicular AP rotations for pose (8%). There was no difference in explained pose variation whether using the humerus, scapula, or clavicle as predictors. However, pose variations were more intertwined between the scapula and the clavicle when using all three shoulder bones as predictors compared to single-bone models, suggesting a stronger coupling between these bones that likely originates from their shared anatomical constraints with the thoracic cage. The dataset showed a mild to excellent fit to the analysis (R² = 0.4-0.9); however, the findings lacked generalizability (Q² = 0.0-0.1), suggesting that PLSRs require additional information-such as soft tissue contributions-to enhance predictive performance. These findings have potential clinical applications in surgical planning, where individual bone shape could be used to estimate native shoulder poses in trauma patients; however, stronger predictive models incorporating additional anatomical or biomechanical parameters are needed to support such use.