Population-specific statistical shape modeling of the Japanese shoulder: a combined analysis of the scapula and humerus stratified by sex and stature.

Background: The smaller shoulder morphology typical of East Asian, especially Japanese, populations results in challenges when using implants designed from North American and European anatomies. Previous Japanese morphologic studies have focused on average shapes without adequately capturing population variability. We aimed to develop a combined statistical shape model (SSM) of the scapula and humerus reflecting covariation, and to quantify subgroup differences by sex and stature with corresponding point clouds. Methods: Pre-operative computed tomography scans from 114 Japanese arthroplasty patients without significant deformity were segmented, mirrored to right-sided geometry where applicable, and remeshed. Correspondence points were optimized separately and connected to create a combined SSM. Principal component (PC) analysis was performed, with both visualization of covariation at ± 3 standard deviations of each mode and calculation of morphological parameters at the same intervals. We evaluated the model performance for compactness (explained variation), generalization (leave-one-out cross-validation), and specificity using 10,000 random shapes. Morphological parameters were measured with SSM-derived correspondence points and compared across 6 sex- and stature-based subgroups (eg, female-140 to male-170). Results: In the combined SSM, PC1 represented global scaling, PC2 reflected coracoid variation, and PC3 represented scapular aspect ratio and lateral scapular structures. Compactness indicated that 95% of the variance was explained by 13 PCs. Regarding generalization, leave-one-out cross-validation error decreased as PCs and sample size increased, converging 26 PCs and 22 cases. Specificity testing showed that even as the number of PCs increased, the 10,000 generated shapes consistently showed low average point distance to the nearest training samples. The mean ± standard deviation values of parameters were similar between the subject and randomly generated groups. Cross-bone prediction with the combined SSM yielded lower errors than baseline mean-shape estimates. Subgroup analyses demonstrated significant differences of the morphologic parameters by sex and stature, including scapula height (132 ± 8 mm in female-140 to 165 ± 7 mm in male-170, Conclusion: A combined Japanese shoulder SSM was developed that captured the morphological variance with high model performance. Corresponding points enabled precise characterization of subgroup-specific morphologies stratified by sex and stature. This stratified model may be utilized for patient-specific treatment, including implant design and positioning.